Raptors|Inroads to Multifingered Grasping - Semantic Scholar
Raptors|Inroads to Multi ngered Grasping Ann M. Ramos and Ian D. Walker Electrical and Computer Engineering, Clemson University, Clemson, SC 29634 [email protected], [email protected]
Abstract
chanical hand seems out of current technology's reach [5].
In this work, we consider the grasping and manipulation strategies of raptors, focusing on the particularly successful case of the osprey. The osprey makes superb use of its two four-digit feet, each of which has ve degrees of freedom. Its manipulation strategies exploit not only quasistatic but also dynamic grasping, particularly in shing, for which the bird is highly renowned. In this paper, we investigate the unusual kinematic design of the osprey foot, and consider the capabilities of the foot for achieving stable grasps. We also perform a dynamic analysis of the osprey in shing, where the talons dynamically impact the sh. Implications for robot hand design and dynamic grasping are discussed. 1
Introduction
Figure 1: Perched osprey holding a sh in one foot [1]
Looking at the world from a human perspective often means perceiving the workspace as an inherently human one. Consequently, the human hand has become the model of choice for robotists when conceptualizing end e ectors for use in a human workspace. We consider ourselves to be the premier species on the planet in no small part due to the advantages a orded us over the lesser creatures by our more dexterous hand.
This raises the natural question: why not consider some of the alternatives while trying to further our understanding of multi ngered manipulation? Perhaps we can learn from the success of other members of the animal kingdom. This is not a novel concept, for example see [13]. However, there may be much to gain in examining successful grasping behavior with hand geometries which are quite foreign when compared to our own.
Until recently, the anthropomorphic model seems to have been the most acclaimed model at the proverbial table [13]. Most multi ngered robot hand designs are based on the anthropomorphic design of a single opposable \thumb" with two or more simpler ngers [5],[9],[14].Yet there are many examples of creatures using hands of di ering geometries (typically of much simpler design than human hands) successfully throughout the planet. One need only visit the local zoo to see a myriad of alternatives to explore.
Consider the grace and power of predatory birds, or raptors. Many of us have been awed to watch a raptor hover almost motionlessly over its prey only to suddenly drop its wings and dive upon its next meal, seemingly e ortlessly catching the prey up in its erce talons. Clearly these birds are superb hunters, but they must also possess manipulators capable of capturing and constraining their writhing prey to be able to consume what they have caught. To focus this line of thought, we chose to center our initial exploration on a particular species of predatory bird{the osprey, Pandion haliaetus (Figure 1).
At this time, a replica of the human hand remains elusive|Luke Skywalker's hand has yet to be built [7]. Even if one could build the physical device itself, the science to control and power it is lacking. With its 27 degrees of freedom, assortment of sensors, and complicated control system, the truly anthropomorphic me-
The osprey, or sh hawk, is such a proli c creature that it is found on all the continents except Antarctica [10]. It can be seen along many of New England's coastlines 1
2.2 Grasp Stability Analysis
each spring as it hunts surface-dwelling sh alongside human shermen. According to some estimates, ospreys successfully catch and keep their prey seventy ve percent of the time, making them some of nature's most successful hunters [3]. In much of the ornithology literature, this success has been attributed to the unique design of the osprey foot. Their feet seem to be designed speci cally for shing, with barbous scales lining their toes, sharply curved talons, studded toe pads, and an abductable fourth toe which can swing around to act in opposition to the two remaining toes [12] (see Figure 2). This uniquely actuated fourth toe is a feature seen only in ospreys and some owls and is attributed with these birds superior hunting skills [10]. It was this feature that initially drew our attention to these magni cent creatures.
A key issue in the analysis of any potential dextrous end e ector design is that of grasp stability. Grasp stability has been de ned in a variety of ways [14], but is generally de ned to be the ability of the end e ector to maintain its grasp of a held object under various classes of contact conditions and disturbances. A common test for general grasp stability involves the notion of form closure, which is not attainable using a parallel jaw gripper. Form closure, or complete restraint, is achieved by any grasp which prevents motion of the object using only unilateral, frictionless contacts [9]. This type of analysis relies very heavily upon having adequate information about the object with which one wishes the manipulator to interact.
In addition to being skillful hunters, ospreys construct nests of more than six feet in diameter, weighing up to one half ton|an impressive feat given that they generally weigh only 2{4 pounds themselves [10]. It seems as though nature is doing something quite unusual in these birds and manipulation may very well be at the heart of it. 2
End e ectors which fail to achieve form closure may indeed, however, have great utility which is not illustrated by considering only complete restraint. An alternative measure, force closure, is achieved by any grasp in which motions of the object are constrained by contact forces [9]. Since it explicitly considers the mechanical advantage of friction, considering whether or not a grasp achieves force closure may elucidate the grasp's adequacy for the task at hand. In this paper, we will analyze the force closure properties of the osprey foot.
Background
To fully appreciate the utility of the ospreys' manipulators, a good understanding of some of the methods and models commonly used in multi ngered hand analysis is helpful. The following section contains a summary of the key methods that will be employed in the remainder of this paper.
Conceptually, given the appropriate degrees of freedom (DOF), a multi ngered hand should be able to achieve stability in some of the cases where a parallel jaw gripper will not suce. Multi ngered hands have the advantage of being able to distribute the payload over several ngers instead of just through two. The addition of articulated joints to the ngers of a hand should provide for greater stability as whole ngers are allowed to conform to an object, forming a power grasp. This advantage, while interesting, will be neglected for the remainder of this work as only ngertip grasps are considered in our analysis of the raptor manipulator.
2.1 Multi ngered End E ectors For years, the robotics community has been striving toward a better end e ector than what is currently in use, parallel jaw grippers. Over the last few years, numerous multi ngered hands have been designed, constructed, and tested [14]. Many of these hands have featured extensive degrees of freedom, with three, four, or even ve ngers. Most designs have been based on the human design of a single `thumb' opposing two or more ngers. However, few of these hands have found their way to practical application. A key problem in this area is the complexity of the designs, which has resulted in relatively low reliability and high control complexity [14].
2.3 Dextrous Manipulation In addition to the ability to stably grasp an object, clearly a desired characteristic of a multi ngered hand is the ability to dextrously manipulate the object. Currently, most analysis of manipulation for robot hands has been based on quasistatic models [14], i.e. solving a static problem at each instant of time. There is relatively little analysis in the robot hand literature which includes hand ( nger) dynamics, and still less of dynamic interactive manipulation itself, although there has been some recent signi cant work in this area [6]. However, as we will see, the osprey uses its feet in a highly dynamic manner, involving sudden contact
In this paper, we consider a relatively simple multi ngered end e ector, the foot of the osprey. To analyze the capabilities of the design, we draw on several techniques commonly in use for multi ngered robot hand design, as discussed in the following section. 2
Figure 2: (A) Close-up view of osprey foot. (B) Schematic representation. and impact with the environment. Thus a dynamic impact model such as the one discussed in [15] will be used in the analysis that follows. We begin the analysis with a description and discussion of the osprey foot, and the way the bird uses it in grasping and manipulation. 3
Raptors
Ospreys seem to be quite capable manipulators and are perceived to be more dextrous than other predatory birds. In the following section, we consider the kinematics of the osprey foot and its use in shing. This discussion is followed by analyses of the osprey's foot as a manipulator.
Figure 3: (A) Zygodactyl foot schematic. (B) Anisodactyl foot schematic. or synchronously an estimated 90 degrees. It seems as though any abducting motion present in toes 2 and 3 results from bone geometry and does not arise from separate actuation. These motions combine to give the osprey a total of 5 DOF per foot.
3.1 Kinematics of the Raptor Foot The osprey has a four-toed foot, the rst and fourth toes of which may be in opposition to the remaining toes (Figure 2). The rst toe, or hallux, is functional in most birds and is always in opposition to the remaining three toes. In owls and ospreys, the fourth toe can be abducted through at least 45 degrees so that it, too, may be opposed such that these birds have both zygodactyl (Figure 3A) and anisodactyl traits (Figure 3B). The number of joints per toe corresponds to the toe number [12]. The most distal phalange on each toe is a sharply curved non-retractable talon. The toes each move as a single unit, i.e. the joints along each toe are coupled such that the bird does not have control over individual joints but instead controls the exion and extension of entire toes. Toes 2{4 have musculature which enables them to be exed either independently
An added advantage of the osprey foot must be accounted for before one can truly appreciate the whole picture. The entire foot is covered with barbs, or scales, which greatly contribute to the osprey's ability to maintain its grip on a slippery sh by significantly increasing the coecient of friction between the sh and its captor's foot. The talons themselves, which sink into the prey's esh in one uid motion, are capable of resisting the writhing, twisting action of a sh trying to free itself. This translates into an ability of the toe to maintain contact even during instances where pulling forces, forces which would otherwise bring about a loss of contact, occur. Thus, each toe is capable of exerting bilateral forces normal to 3
rotate its prey during ight while maintaining a secure grasp. This seems possible due to toe redundancy in a given grasp. To more clearly see how this might be true, consider the grasp in Figure 4. The forward placement of the grasp with respect to the center of mass is due to the bird grasping the sh near its head, minimizing the sh's ability to twist about the z axis forward of the grasp. The direction of ight is in the ;y direction. For the purposes of notation, f = [fX f Y f Z ]T ; t = [tX tY tZ ]T ; X = ff jf Y ; f Z = 0g; +X = ff jfY; f Z = 0; fX > 0g; and tX is the moment about the x axis where +tX = ftjtY ; tZ = 0; tX counter-clockwiseg. For the grasp in Figure 4, we consider four distinct contact cases and examine the rami cations of each in achieving grasp stability and manipulation.
the contact point when piercing occurs. In these situations, each toe can both push and, to a lesser degree, pull. The resulting friction cone is suciently large to allow a single contact point to constrain translations in the x, y, and z directions. The ability to resist moments varies with the type of contact, as will be seen in the next section. A digitigrade, like all birds, the osprey walks upon its toes. In ight, the feet are kept tightly closed against the body, minimizing wind resistance. The feet remain in this position until just before landing or hitting the surface of the water when hunting [10]. In either case, the toes are oriented in zygodactyl fashion and the legs are outstretched such that the bird appears to be diving feet rst. When shing, the osprey enters the water, grasps its prey in one or both feet (depending upon its target's size and weight) with at least one talon piercing the sh, and then re-emerges from the water and takes to the air. If the bird does not have a solid grasp of the sh, it may regrasp while in ight. In some cases, the osprey may decide to re-orient the sh to reduce drag or improve its ability to hold onto its slippery prey. Watching an osprey catch, carry, and kill a sh weighing up to 30% of its body weight is quite impressive [8].
3.2 Analysis of Raptor Foot Design To further our understanding of the workings of the osprey foot, we next consider a stereotypical grasp which the birds have been observed to use while shing. This by no means represents a complete view of osprey shing strategy. As mentioned earlier, ospreys may sometimes choose to use both feet to constrain a large sh or may use only a couple of talons in gripping a sh. In any case, we have come to see the osprey foot as a set of recon gurable hooks (talons) which are used to pick up, constrain, and kill its prey.
Figure 5: The contact cases considered for raptor manipulation can be viewed as acting like Mason and Salisbury's known contact schemes: (A) Planar contact with friction. (B) Line contact with friction. (C) Soft nger. (D) Point contact with friction. [9]
3.2.1 Case 1: Full talon piercing For this rst case, we consider the situation where a talon fully punctures the object which the raptor is grasping. As a result of the talon's curvature and depth of contact, each toe1 resists all translations and all moments, Figure 5(A). Additionally, each toe is able to resist bilateral forces|both pushing and
Figure 4: Stereotypical grasp. Numbers indicate the position of the various toes. The following static analysis considers only \ ngertip" grasping, that is we shall neglect the use of the studded foot pad in the grasp considered. Observations indicate the ability of the bird to both lift and
1 For the remainder of this work, the terms nger and toe are considered interchangeable as they pertain to end e ectors.
4
sary for grasp stability, the degree of redundancy (and thus available DOFs for manipulation), and what combinations of ngers might be more advantageous from the regrasping point of view, consider Table 2.
pulling motions of its prey. Grasps which been observed include one-footed grasps of the prey on only a single plane, validating our conclusion that the talon is capable of resisting pulling forces. As we have seen, force closure can be achieved by a single toe when its talon completely pierces the object it is grasping.
As can be seen, combinations of toes produce moments that neither toe could produce independently as a result of the moment arm between them. Despite this, no combination of two toes is able to achieve force closure for the given contact points. All the three- ngered grasps are, however, able to attain force closure, giving the remaining toe the freedom to be involved in regrasping or to otherwise put its redundancy into use. The greatest freedom results from the rst three ngered combination, Finger 1, Finger 2, & Finger 3, because the osprey then has 2 extra DOF.
3.2.2 Case 2: Tip piercing Next, we consider the situation in which only the tip of the talon pierces the object of interest. Because only the tip is involved, the talon acts only along the z axis in Figure 5(B). As in the preceding case, each talon resists all translations. However, under these circumstances, a talon can only resist moments about the x and y axes. Given the DOFs each toe is able to constrain, tip piercing is equivalent to line contact with friction. Additionally, since piercing occurs, the toes are capable of both pushing and pulling. Temporarily ignoring pulling for the grasp pictured in Figure 4, each toe is able to resist motions in the following directions:
3.2.4 Case 4: Nest building
Finger 1: X [ Y[ + Z [ tX [ tY Finger 2: X [ Y [ Z [ tY [ tZ Finger 3: X [ ;Y [ Z [ tX [ tZ Finger 4: ;X [ Y [ Z [ tY [ tZ
If we consider the grasp pictured in Figure 4 for a multi ngered hand given point contacts with friction, the situation is somewhat less advantageous. This case corresponds to a raptor manipulating a rigid object with only its \ ngertips." From Figure 5(D), each talon resists all translations but can only resist pushing forces in the z direction. Using the toe con guration in Figure 4 with four point contacts, we see that the grasp is able to constrain X [ Y [ Z [ +tX [ tY [ tZ but is unable to generate a clockwise moment about the x axis (;tX ). Thus, a fth toe is necessary in one of two places (see Figure 6) to achieve force closure. Thus, given only point contacts with friction, the osprey is unable to achieve force closure using only a single foot. In these instances, it would have to perform the task in question using an additional surface (such as the ground or side of the nest) or by using both of its feet.
Looking at combinations of toes in Table 1, we see that complete restraint can be achieved by any two toes as long as those toes do not lie exactly parallel to one another. As it is not very likely for this condition to occur, force closure can usually be attained by any two toes in grasps which involve only tip piercing. The situation does not change if pulling forces are allowed to occur.
3.2.3 Case 3: Compliant grasping Of course, there are instances when piercing the object an osprey holds is not possible or even ideal. In these situations, the surface of the object is either rigid (examined in Case 4) or compliant. Given a compliant surface, the toes can be considered to act as soft ngers. Therefore, each toe can resist translations in the x, y, and +z directions and moments about z, see Figure 5(C). For the grasp pictured in Figure 4, each toe is capable of resisting motions in the following directions: Finger 1: X [ Y[ + Z [ tZ Finger 2: X [ Y [ Z [ tX Finger 3: X [ ;Y [ Z [ tY Finger 4: ;X [ Y [ Z [ tX
Given the four- ngered grasp originally considered for the osprey, one way to achieve form closure is to add both of the 3 ngers mentioned above to the grasp, thus spanning < in the frictionless case. Without this addition, none of the contact strategies discussed attain form closure. While this analysis follows the conventions for stability we originally introduced, it is unsatisfying as it seems to deem the osprey's grasp unstable (because it always lacks form closure in the singlefooted case) yet, given the osprey's success, this designation doesn't appear correct. Considering the static frictionless case as an indication of stability seems overly stringent and contrived. We next consider the implications of introducing dynamics into our evaluation of raptorial manipulation.
To gain an understanding of the number of toes neces5
Toe 1p p2 3 p 1 p1 p 1 1 p 1 p p 1 1 1 1 p
4 Implications 2 Force Closure Closure p2 Force Force Closure Closure p2 Force lacking tX p Force Closure
Table 1: Check marks indicate those toes involved in grasping. The number in the rst 4 cells of each grasp indicates the remaining DOFs which can be used for regrasping, etc.
Toe 1p 2p 3 p 1 1p p 1 1 p 1 p p 1 1 1p 1p p p
Constrained 4 Directions Implications 2 X [ Y [ Z [ tX [ +tY [ tZ lacking ;tY 2p X [ Y [ Z [ +tX [ tY [ tZ lacking ;tX X [ Y [ Z [ tX [ ;tY [ tZ lacking +tY 2p X [ Y [ Z [ tX [ tY [ +tZ lacking ;tZ [ Y [ Z [ tX lacking tY ; tZ p X X [ Y [ Z [ tX [ tY [ ;tZ lacking +tZ [ Y [ Z [ tX [ tY [ tZ Force Closure p p 1 2p X X [ Z [ tX [ tY [ tZ Force Closure p 1 p p X [[ Y Y [ Z [ tX [ tY [ tZ Force Closure 1 p p p X [ Y [ Z [ tX [ tY [ tZ Force Closure Table 2: Check marks indicate those toes involved in grasping. The numbers in the rst 4 cells of each grasp indicates the remaining DOFs which can be used for regrasping, etc should the grasp achieve force closure.
Figure 6: (A) and (B) are two possible ways to achieve force closure using only point contacts.
6
3.3 Dynamic Grasping From the previous section, we have seen that in order to achieve force closure of a grasped object without two-footed grasping or environmental constraints, some piercing of the object is necessary. This restriction is, of course, due to the relative simplicity of the foot design. Therefore it is not surprising that the osprey, in hunting and shing, opts for the highly dynamic approach of spearing its prey with the talons of one or both of its feet. This mode of manipulation, which clearly involves dynamic impact at the end e ector (here the talons), is one that is receiving increased attention by robotics researchers. This interest is due to the fact that many manipulation tasks can be completely transformed when approached from a dynamic contact (as opposed to quasistatic contact, the traditional approach in robotics) viewpoint. Many tasks (for example extracting a sh from water) can be performed much faster and more eciently with impulsive manipulation [6]. It has been previously noted that many animals, such as raccoons [13] appear to use dynamic manipulation modes extensively in nature, making good use of their relatively simple end e ectors.
Figure 7: Model of osprey foot used in dynamic analysis. sh from behind. As discussed earlier, this appears to be the case that the osprey follows in practice [8], and no doubt has the added advantage of allowing the bird to track the sh more accurately during descent. The outstretched con guration of the talons preferred by the osprey at the instant of impact minimizes the impact forces imparted to the toe joints.
To investigate the dynamic grasping behavior of the osprey, a dynamic simulation of its shing behavior was conducted. The osprey foot kinematics (see Figure 2) were approximated by the model in Figure 7 (where in both cases the shaded region denotes coupling between the displayed joints, and the joints connected by a shaded region represent one independent degree of freedom). Average mass and dimensional parameters for the osprey and sh were taken from [3] and [16]. For the moment of contact, the robot impact model in [15] was used. The sh was assumed to be a point mass, and the contact conditions to correspond to a plastic collision. The drag induced by the bird hitting the water was neglected for simplicity (in essence, the sh is assumed to be immediately below the surface of the water).
The impact force is also strongly a function of the normal to the plane of contact, which in this case corresponds to the geometry of the sh. The contact forces drop signi cantly (an order of magnitude) depending on whether the strike plane is parallel to the water surface (the highest value, corresponding to hitting the sh along its back), or more perpendicular to the water surface (much lower values corresponding to hitting the sh along the downward slope of its head or side). Interestingly, a combination of a dive angle of 45 degrees (the observed typical dive angle [8]) and a contact normal aligned with the dive angle result in an impact force from the model independent of the velocity of the bird! This is due to the fact that for rigid body collisions, the impulsive force is a function of the bodies' pre-collision velocity normal to the plane of contact [15]. In practice this implies that, by matching its dive angle to the plane of contact on the sh, the bird can achieve essentially the same impact (capturing) force on the sh over a wide range of velocities. Of course, there are practical constraints on the velocity of the bird due to the need to track the sh in the water before impact. However, this inherent possibility of achieving the same force task goal over a variety of closing contact velocities and con gurations with dynamic manipulation is intriguing when considered from the robotics point of view.
The results of the simulation con rm the e ectiveness of the osprey shing strategy. For the case of single talon contact (talons 1 and 3 were considered), the contact condition that resulted in the highest impact force imparted to the sh, (for the typical raptor closing velocity of 13:4 m=s) was when the raptor descended along the opposite direction to the sh trajectory. This is to be expected, and corresponds to a `head-on' collision, resulting in a high impulsive force imparted to the talons, which may be undesirable to the bird. Much smaller impact forces to the sh (of at least a factor of two) result from the opposite case, with the osprey following the sh trajectory, and hitting the
Thus, the results of the dynamic simulation of osprey 7
up to ve feet in length [8]. To bring these building supplies to the nest site, an osprey must y while holding the item in both feet, using only point contacts with friction. During landing, the bird releases one foot from its cargo and uses it to guide the twigs, through a series of impulses, into the correct position by using the foot still grasping as a pivot point. Once the nest's supporting structure has been built, a soft bed of materials suitable for eggs must be gathered. Osprey nests have been found which contain a large variety of trash comprising this layer. Grass, shing line, seaweed, and rope have all been reported [8], [10]. Because these materials are soft, we surmise they were brought to the nest by compliant grasping. This accounts for observations that an osprey may hold these materials in a single foot during ight. As we have seen, the formalism we have applied in examining the various contact situations seems to agree well with our intuition into everyday raptor grasping.
shing yield information which supports and provides new understanding of the osprey hunting behavior. In particular, the dynamic grasping strategy clearly compensates for the lack of kinematic dexterity in the hand. In addition, the shing example provides some insights into how dynamic manipulation with simple hands can be e ectively conducted. 4
Discussion
Having well established a perspective on raptorial contact situations and having examined a stereotypical dynamic grasping situation, we now consider the implications of our analyses on other raptor grasping tasks. Consider rst the task of eating. After having caught a sh, an osprey may choose to perch in a nearby tree to consume its prey. To do so, it must support its weight with one foot and hold the sh in the other. If the sh is fairly large compared to the size of the foot, the osprey will not pierce the sh with its talons, but will instead use a compliant grasp, encircling its food as much as possible to prevent it from getting away. While maintaining its balance on the other foot, it will lift the sh to its mouth and tear a bite-size piece o with its beak, e ectively introducing an impulse to the system which the grasp has to resist. This process is repeated until the remainder of the sh is about the size of the bird's foot. At this point, constraining the food with the whole foot encumbers the osprey's ability to consume the remaining meat. Thus, it must switch to a di erent contact regime. Our second contact case, tip piercing, takes over and the bird pierces its food with the tips of one or two talons. Constraining the food in this manner, our osprey consumes all but the tail and toughest of scales, allowing little to go to waste [10].
The kinematic analysis suggests that the osprey hand is quite limited. However, as we have seen, the osprey is able to display highly successful specialized grasping behaviors. Obviously, the human hand is signi cantly more versatile than the osprey foot. However, this is due to its signi cantly increased complexity, which we are not close to reproducing with robotic devices. The simpler design of the osprey foot presents a more achievable design with current technology. Thus, an interesting question that arises from the analysis in this paper is: are there aspects of the grasping behavior of the osprey which can be successfully used in the robotics arena? Several points arise in this regard. First, the use of the talons for piercing is clearly a key to overcoming the diculties in achieving force closure for the osprey. The use of `claws' on end e ectors is not an area that has received signi cant attention in robotics. Clearly, many applications preclude the use of talons or claws, due the the need to protect the environment or payload. However, for some applications this may represent a useful grasping mode.
When a male osprey is a member of a breeding pair, his chief responsibility is to provide food to his mate and young. Taking his share of the meal rst, the male osprey will consume the head of its prey as described above and deliver the rest to its mate and their young where the female will feed the young. In this context, the female is able to use only point contacts with friction (Case 4 above) to constrain the meal since it will o er her no resistance. She can achieve force closure by holding the sh against the oor of the nest. The osprey will tear o small pieces of meat with her beak and feed them to her hungry young. Other than the contact type employed, the rest of the process is the same as for general eating which was discussed earlier.
Secondly, the use of the environment to provide the `extra' ngers to achieve force closure is interesting. Almost all robots operate independently of the world as much as possible, attempting to achieve the entire task within the robot structure. The use of `props' in the world surrounding the robot is clearly a successful one for the osprey, as well as for many other creatures. Finally, the strong and successful use of dynamic manpulation by the osprey reveals how, by interacting dynamically with the world as opposed to the quasistatic mode almost universal in robotics, allows some tasks to be performed more eciently. In fact, in some cases, the task could not be performed at all without the dynamic manipulation mode. It appears that the issue of
Finally, let us consider the task of nest building. A young breeding pair may spend their entire rst season together constructing a nest. First, they build a frame for the nest in a (preferably dead) tree out of sticks and twigs. These rst components may be 8
dynamic manipulation exempli ed by the osprey can be a key mode in the future for robotics. 5
References [1] Corel Photo Studio. Image Number 591005 in Corel Photo CD Collection, 1997. [2] Cutkosky, M.R. Robotic Grasping and Fine Manipulation. Kluwer Academic Publishers, Boston, 1985. [3] Dunne, P. The Wind Masters: The Lives of North American Birds of Prey. Houghton Miin Company, Boston, 1995. [4] Gill, F.B. Ornithology. Second Edition. W.H. Freeman and Company, New York, 1995. [5] Grupen, R.A., Henderson, T.C. and I.D. McCammon. A Survey of General Purpose Manipulators. In The International Journal of Robotics Research, Vol. 8, No. 1, pages 38-62, 1989. [6] Huang, H.H., Krotkov, E.P., and M.T. Mason. Impulsive Manipulation. In Proceedings 1995 IEEE International Conference on Robotics and Automation, pages 120-125, Nagoya, Japan, 1995. [7] Lucas, G. The Empire Strikes Back. LucasFilm Ltd. Released by Twentieth Century Fox, 1980, 1997. [8] Male, M. and J. Fieth. Nova: Return of the Osprey. Blue Earth Films for WGBH Educational Foundation, 1986. [9] Mason, M.T. and J.K. Salisbury. Robot Hands and the Mechanics of Manipulation. MIT Press, Cambridge, Massachusetts, 1985. [10] Poole, A.F. Ospreys: Their Natural and Unnatural History. Cambridge University Press, Cambridge, United Kingdom, 1989. [11] Proctor, N.S. and P.J. Lynch. Manual of Ornithology: Avian Structure and Function. Yale University Press, New Haven, Conneticut, 1993. [12] Van Tyne, J. and A.J. Berger. Fundamentals of Ornithology. Second Edition. John Wiley & Sons, New York, 1976. [13] Walker, I.D. A Successful Multi ngered Hand Design|The Case of the Raccoon. In Proceedings 1995 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 186-193, Pittsburg, PA, 1995. [14] Walker, I.D. Modeling and Control of Mult ngered Hands: A Survey. To appear in Complex Robotic Systems, P. Chiacchio and S. Chiaverini, eds., Lecture Notes in Control and Information Sciences Series, Springer-Verlag, London, 1998. [15] Walker, I.D. Impact Con gurations and Measures for Kinematically Redundant and Multiple Armed Robot Systems. In IEEE Transactions on Robotics and Automation, Vol. 10, No. 5, pages 670-683, 1994. [16] Wheeler, B.K. and W.S. Clark. A Photographic Guide to North American Raptors. Academic Press, London, 1995.
Conclusions
In this paper, we have considered a highly successful example of a multi ngered end e ector in nature, namely the four-toed osprey foot. The osprey foot is of an unusual kinematic design, and the bird uses its feet in novel ways compared to the typical use of dextrous robotic end e ectors. We have used the concept of form and force closure to analyze the capability of the osprey foots kinematic design to achieve grasp stability. In addition, a dynamic analysis of the osprey in shing mode, in which the bird dives and dynamically captures the sh, has been conducted. The results of the analyses con rm several key physical factors underlying the grasping and manipulation strategy of the osprey. For quasistatic grasping situations, the bird's use of the environment to achieve grasp stability seems a necessary result of the relative simplicity of the foot kinematics. From the dynamic analysis one can see how the bird optimizes its ight trajectory and toe con guration to best achieve dynamic impact with the target when shing. From the perspective of robotics, the design of the osprey foot is appealing due to its relative simplicity and compactness. Robot hands based on the osprey design would be simpler and cheaper than those based on the human design. This could result in signi cant bene ts in robotic applications if the functionality of the hands was sucient for application tasks. A key issue here is the inherently dynamic way in which the osprey uses its feet in many manipulation tasks. The use of dynamic manipulation, which inherently implies dynamic impact, allows tasks to be performed faster and in many cases (for example osprey shing) more e ectively than using a quasistatic approach. It seems that many creatures with relatively simple multi ngered end e ectors, such as the osprey, successfully use dynamic manipulation to compensate for the lack of dexterity in their hands. This provides motivation for the robotics community to further investigate the issue of how the use of dynamic manipulation could allow many tasks to be undertaken in a more e ective manner, by simpler robot hand designs than those currently under investigation. This work was supported under NSF grant number CMS9796328, by the U.S. Department of Energy under contract number DE-FG07-97ER14830, and a grant from the Clemson University Research Grant Committee.
9
Abstract
chanical hand seems out of current technology's reach [5].
In this work, we consider the grasping and manipulation strategies of raptors, focusing on the particularly successful case of the osprey. The osprey makes superb use of its two four-digit feet, each of which has ve degrees of freedom. Its manipulation strategies exploit not only quasistatic but also dynamic grasping, particularly in shing, for which the bird is highly renowned. In this paper, we investigate the unusual kinematic design of the osprey foot, and consider the capabilities of the foot for achieving stable grasps. We also perform a dynamic analysis of the osprey in shing, where the talons dynamically impact the sh. Implications for robot hand design and dynamic grasping are discussed. 1
Introduction
Figure 1: Perched osprey holding a sh in one foot [1]
Looking at the world from a human perspective often means perceiving the workspace as an inherently human one. Consequently, the human hand has become the model of choice for robotists when conceptualizing end e ectors for use in a human workspace. We consider ourselves to be the premier species on the planet in no small part due to the advantages a orded us over the lesser creatures by our more dexterous hand.
This raises the natural question: why not consider some of the alternatives while trying to further our understanding of multi ngered manipulation? Perhaps we can learn from the success of other members of the animal kingdom. This is not a novel concept, for example see [13]. However, there may be much to gain in examining successful grasping behavior with hand geometries which are quite foreign when compared to our own.
Until recently, the anthropomorphic model seems to have been the most acclaimed model at the proverbial table [13]. Most multi ngered robot hand designs are based on the anthropomorphic design of a single opposable \thumb" with two or more simpler ngers [5],[9],[14].Yet there are many examples of creatures using hands of di ering geometries (typically of much simpler design than human hands) successfully throughout the planet. One need only visit the local zoo to see a myriad of alternatives to explore.
Consider the grace and power of predatory birds, or raptors. Many of us have been awed to watch a raptor hover almost motionlessly over its prey only to suddenly drop its wings and dive upon its next meal, seemingly e ortlessly catching the prey up in its erce talons. Clearly these birds are superb hunters, but they must also possess manipulators capable of capturing and constraining their writhing prey to be able to consume what they have caught. To focus this line of thought, we chose to center our initial exploration on a particular species of predatory bird{the osprey, Pandion haliaetus (Figure 1).
At this time, a replica of the human hand remains elusive|Luke Skywalker's hand has yet to be built [7]. Even if one could build the physical device itself, the science to control and power it is lacking. With its 27 degrees of freedom, assortment of sensors, and complicated control system, the truly anthropomorphic me-
The osprey, or sh hawk, is such a proli c creature that it is found on all the continents except Antarctica [10]. It can be seen along many of New England's coastlines 1
2.2 Grasp Stability Analysis
each spring as it hunts surface-dwelling sh alongside human shermen. According to some estimates, ospreys successfully catch and keep their prey seventy ve percent of the time, making them some of nature's most successful hunters [3]. In much of the ornithology literature, this success has been attributed to the unique design of the osprey foot. Their feet seem to be designed speci cally for shing, with barbous scales lining their toes, sharply curved talons, studded toe pads, and an abductable fourth toe which can swing around to act in opposition to the two remaining toes [12] (see Figure 2). This uniquely actuated fourth toe is a feature seen only in ospreys and some owls and is attributed with these birds superior hunting skills [10]. It was this feature that initially drew our attention to these magni cent creatures.
A key issue in the analysis of any potential dextrous end e ector design is that of grasp stability. Grasp stability has been de ned in a variety of ways [14], but is generally de ned to be the ability of the end e ector to maintain its grasp of a held object under various classes of contact conditions and disturbances. A common test for general grasp stability involves the notion of form closure, which is not attainable using a parallel jaw gripper. Form closure, or complete restraint, is achieved by any grasp which prevents motion of the object using only unilateral, frictionless contacts [9]. This type of analysis relies very heavily upon having adequate information about the object with which one wishes the manipulator to interact.
In addition to being skillful hunters, ospreys construct nests of more than six feet in diameter, weighing up to one half ton|an impressive feat given that they generally weigh only 2{4 pounds themselves [10]. It seems as though nature is doing something quite unusual in these birds and manipulation may very well be at the heart of it. 2
End e ectors which fail to achieve form closure may indeed, however, have great utility which is not illustrated by considering only complete restraint. An alternative measure, force closure, is achieved by any grasp in which motions of the object are constrained by contact forces [9]. Since it explicitly considers the mechanical advantage of friction, considering whether or not a grasp achieves force closure may elucidate the grasp's adequacy for the task at hand. In this paper, we will analyze the force closure properties of the osprey foot.
Background
To fully appreciate the utility of the ospreys' manipulators, a good understanding of some of the methods and models commonly used in multi ngered hand analysis is helpful. The following section contains a summary of the key methods that will be employed in the remainder of this paper.
Conceptually, given the appropriate degrees of freedom (DOF), a multi ngered hand should be able to achieve stability in some of the cases where a parallel jaw gripper will not suce. Multi ngered hands have the advantage of being able to distribute the payload over several ngers instead of just through two. The addition of articulated joints to the ngers of a hand should provide for greater stability as whole ngers are allowed to conform to an object, forming a power grasp. This advantage, while interesting, will be neglected for the remainder of this work as only ngertip grasps are considered in our analysis of the raptor manipulator.
2.1 Multi ngered End E ectors For years, the robotics community has been striving toward a better end e ector than what is currently in use, parallel jaw grippers. Over the last few years, numerous multi ngered hands have been designed, constructed, and tested [14]. Many of these hands have featured extensive degrees of freedom, with three, four, or even ve ngers. Most designs have been based on the human design of a single `thumb' opposing two or more ngers. However, few of these hands have found their way to practical application. A key problem in this area is the complexity of the designs, which has resulted in relatively low reliability and high control complexity [14].
2.3 Dextrous Manipulation In addition to the ability to stably grasp an object, clearly a desired characteristic of a multi ngered hand is the ability to dextrously manipulate the object. Currently, most analysis of manipulation for robot hands has been based on quasistatic models [14], i.e. solving a static problem at each instant of time. There is relatively little analysis in the robot hand literature which includes hand ( nger) dynamics, and still less of dynamic interactive manipulation itself, although there has been some recent signi cant work in this area [6]. However, as we will see, the osprey uses its feet in a highly dynamic manner, involving sudden contact
In this paper, we consider a relatively simple multi ngered end e ector, the foot of the osprey. To analyze the capabilities of the design, we draw on several techniques commonly in use for multi ngered robot hand design, as discussed in the following section. 2
Figure 2: (A) Close-up view of osprey foot. (B) Schematic representation. and impact with the environment. Thus a dynamic impact model such as the one discussed in [15] will be used in the analysis that follows. We begin the analysis with a description and discussion of the osprey foot, and the way the bird uses it in grasping and manipulation. 3
Raptors
Ospreys seem to be quite capable manipulators and are perceived to be more dextrous than other predatory birds. In the following section, we consider the kinematics of the osprey foot and its use in shing. This discussion is followed by analyses of the osprey's foot as a manipulator.
Figure 3: (A) Zygodactyl foot schematic. (B) Anisodactyl foot schematic. or synchronously an estimated 90 degrees. It seems as though any abducting motion present in toes 2 and 3 results from bone geometry and does not arise from separate actuation. These motions combine to give the osprey a total of 5 DOF per foot.
3.1 Kinematics of the Raptor Foot The osprey has a four-toed foot, the rst and fourth toes of which may be in opposition to the remaining toes (Figure 2). The rst toe, or hallux, is functional in most birds and is always in opposition to the remaining three toes. In owls and ospreys, the fourth toe can be abducted through at least 45 degrees so that it, too, may be opposed such that these birds have both zygodactyl (Figure 3A) and anisodactyl traits (Figure 3B). The number of joints per toe corresponds to the toe number [12]. The most distal phalange on each toe is a sharply curved non-retractable talon. The toes each move as a single unit, i.e. the joints along each toe are coupled such that the bird does not have control over individual joints but instead controls the exion and extension of entire toes. Toes 2{4 have musculature which enables them to be exed either independently
An added advantage of the osprey foot must be accounted for before one can truly appreciate the whole picture. The entire foot is covered with barbs, or scales, which greatly contribute to the osprey's ability to maintain its grip on a slippery sh by significantly increasing the coecient of friction between the sh and its captor's foot. The talons themselves, which sink into the prey's esh in one uid motion, are capable of resisting the writhing, twisting action of a sh trying to free itself. This translates into an ability of the toe to maintain contact even during instances where pulling forces, forces which would otherwise bring about a loss of contact, occur. Thus, each toe is capable of exerting bilateral forces normal to 3
rotate its prey during ight while maintaining a secure grasp. This seems possible due to toe redundancy in a given grasp. To more clearly see how this might be true, consider the grasp in Figure 4. The forward placement of the grasp with respect to the center of mass is due to the bird grasping the sh near its head, minimizing the sh's ability to twist about the z axis forward of the grasp. The direction of ight is in the ;y direction. For the purposes of notation, f = [fX f Y f Z ]T ; t = [tX tY tZ ]T ; X = ff jf Y ; f Z = 0g; +X = ff jfY; f Z = 0; fX > 0g; and tX is the moment about the x axis where +tX = ftjtY ; tZ = 0; tX counter-clockwiseg. For the grasp in Figure 4, we consider four distinct contact cases and examine the rami cations of each in achieving grasp stability and manipulation.
the contact point when piercing occurs. In these situations, each toe can both push and, to a lesser degree, pull. The resulting friction cone is suciently large to allow a single contact point to constrain translations in the x, y, and z directions. The ability to resist moments varies with the type of contact, as will be seen in the next section. A digitigrade, like all birds, the osprey walks upon its toes. In ight, the feet are kept tightly closed against the body, minimizing wind resistance. The feet remain in this position until just before landing or hitting the surface of the water when hunting [10]. In either case, the toes are oriented in zygodactyl fashion and the legs are outstretched such that the bird appears to be diving feet rst. When shing, the osprey enters the water, grasps its prey in one or both feet (depending upon its target's size and weight) with at least one talon piercing the sh, and then re-emerges from the water and takes to the air. If the bird does not have a solid grasp of the sh, it may regrasp while in ight. In some cases, the osprey may decide to re-orient the sh to reduce drag or improve its ability to hold onto its slippery prey. Watching an osprey catch, carry, and kill a sh weighing up to 30% of its body weight is quite impressive [8].
3.2 Analysis of Raptor Foot Design To further our understanding of the workings of the osprey foot, we next consider a stereotypical grasp which the birds have been observed to use while shing. This by no means represents a complete view of osprey shing strategy. As mentioned earlier, ospreys may sometimes choose to use both feet to constrain a large sh or may use only a couple of talons in gripping a sh. In any case, we have come to see the osprey foot as a set of recon gurable hooks (talons) which are used to pick up, constrain, and kill its prey.
Figure 5: The contact cases considered for raptor manipulation can be viewed as acting like Mason and Salisbury's known contact schemes: (A) Planar contact with friction. (B) Line contact with friction. (C) Soft nger. (D) Point contact with friction. [9]
3.2.1 Case 1: Full talon piercing For this rst case, we consider the situation where a talon fully punctures the object which the raptor is grasping. As a result of the talon's curvature and depth of contact, each toe1 resists all translations and all moments, Figure 5(A). Additionally, each toe is able to resist bilateral forces|both pushing and
Figure 4: Stereotypical grasp. Numbers indicate the position of the various toes. The following static analysis considers only \ ngertip" grasping, that is we shall neglect the use of the studded foot pad in the grasp considered. Observations indicate the ability of the bird to both lift and
1 For the remainder of this work, the terms nger and toe are considered interchangeable as they pertain to end e ectors.
4
sary for grasp stability, the degree of redundancy (and thus available DOFs for manipulation), and what combinations of ngers might be more advantageous from the regrasping point of view, consider Table 2.
pulling motions of its prey. Grasps which been observed include one-footed grasps of the prey on only a single plane, validating our conclusion that the talon is capable of resisting pulling forces. As we have seen, force closure can be achieved by a single toe when its talon completely pierces the object it is grasping.
As can be seen, combinations of toes produce moments that neither toe could produce independently as a result of the moment arm between them. Despite this, no combination of two toes is able to achieve force closure for the given contact points. All the three- ngered grasps are, however, able to attain force closure, giving the remaining toe the freedom to be involved in regrasping or to otherwise put its redundancy into use. The greatest freedom results from the rst three ngered combination, Finger 1, Finger 2, & Finger 3, because the osprey then has 2 extra DOF.
3.2.2 Case 2: Tip piercing Next, we consider the situation in which only the tip of the talon pierces the object of interest. Because only the tip is involved, the talon acts only along the z axis in Figure 5(B). As in the preceding case, each talon resists all translations. However, under these circumstances, a talon can only resist moments about the x and y axes. Given the DOFs each toe is able to constrain, tip piercing is equivalent to line contact with friction. Additionally, since piercing occurs, the toes are capable of both pushing and pulling. Temporarily ignoring pulling for the grasp pictured in Figure 4, each toe is able to resist motions in the following directions:
3.2.4 Case 4: Nest building
Finger 1: X [ Y[ + Z [ tX [ tY Finger 2: X [ Y [ Z [ tY [ tZ Finger 3: X [ ;Y [ Z [ tX [ tZ Finger 4: ;X [ Y [ Z [ tY [ tZ
If we consider the grasp pictured in Figure 4 for a multi ngered hand given point contacts with friction, the situation is somewhat less advantageous. This case corresponds to a raptor manipulating a rigid object with only its \ ngertips." From Figure 5(D), each talon resists all translations but can only resist pushing forces in the z direction. Using the toe con guration in Figure 4 with four point contacts, we see that the grasp is able to constrain X [ Y [ Z [ +tX [ tY [ tZ but is unable to generate a clockwise moment about the x axis (;tX ). Thus, a fth toe is necessary in one of two places (see Figure 6) to achieve force closure. Thus, given only point contacts with friction, the osprey is unable to achieve force closure using only a single foot. In these instances, it would have to perform the task in question using an additional surface (such as the ground or side of the nest) or by using both of its feet.
Looking at combinations of toes in Table 1, we see that complete restraint can be achieved by any two toes as long as those toes do not lie exactly parallel to one another. As it is not very likely for this condition to occur, force closure can usually be attained by any two toes in grasps which involve only tip piercing. The situation does not change if pulling forces are allowed to occur.
3.2.3 Case 3: Compliant grasping Of course, there are instances when piercing the object an osprey holds is not possible or even ideal. In these situations, the surface of the object is either rigid (examined in Case 4) or compliant. Given a compliant surface, the toes can be considered to act as soft ngers. Therefore, each toe can resist translations in the x, y, and +z directions and moments about z, see Figure 5(C). For the grasp pictured in Figure 4, each toe is capable of resisting motions in the following directions: Finger 1: X [ Y[ + Z [ tZ Finger 2: X [ Y [ Z [ tX Finger 3: X [ ;Y [ Z [ tY Finger 4: ;X [ Y [ Z [ tX
Given the four- ngered grasp originally considered for the osprey, one way to achieve form closure is to add both of the 3 ngers mentioned above to the grasp, thus spanning < in the frictionless case. Without this addition, none of the contact strategies discussed attain form closure. While this analysis follows the conventions for stability we originally introduced, it is unsatisfying as it seems to deem the osprey's grasp unstable (because it always lacks form closure in the singlefooted case) yet, given the osprey's success, this designation doesn't appear correct. Considering the static frictionless case as an indication of stability seems overly stringent and contrived. We next consider the implications of introducing dynamics into our evaluation of raptorial manipulation.
To gain an understanding of the number of toes neces5
Toe 1p p2 3 p 1 p1 p 1 1 p 1 p p 1 1 1 1 p
4 Implications 2 Force Closure Closure p2 Force Force Closure Closure p2 Force lacking tX p Force Closure
Table 1: Check marks indicate those toes involved in grasping. The number in the rst 4 cells of each grasp indicates the remaining DOFs which can be used for regrasping, etc.
Toe 1p 2p 3 p 1 1p p 1 1 p 1 p p 1 1 1p 1p p p
Constrained 4 Directions Implications 2 X [ Y [ Z [ tX [ +tY [ tZ lacking ;tY 2p X [ Y [ Z [ +tX [ tY [ tZ lacking ;tX X [ Y [ Z [ tX [ ;tY [ tZ lacking +tY 2p X [ Y [ Z [ tX [ tY [ +tZ lacking ;tZ [ Y [ Z [ tX lacking tY ; tZ p X X [ Y [ Z [ tX [ tY [ ;tZ lacking +tZ [ Y [ Z [ tX [ tY [ tZ Force Closure p p 1 2p X X [ Z [ tX [ tY [ tZ Force Closure p 1 p p X [[ Y Y [ Z [ tX [ tY [ tZ Force Closure 1 p p p X [ Y [ Z [ tX [ tY [ tZ Force Closure Table 2: Check marks indicate those toes involved in grasping. The numbers in the rst 4 cells of each grasp indicates the remaining DOFs which can be used for regrasping, etc should the grasp achieve force closure.
Figure 6: (A) and (B) are two possible ways to achieve force closure using only point contacts.
6
3.3 Dynamic Grasping From the previous section, we have seen that in order to achieve force closure of a grasped object without two-footed grasping or environmental constraints, some piercing of the object is necessary. This restriction is, of course, due to the relative simplicity of the foot design. Therefore it is not surprising that the osprey, in hunting and shing, opts for the highly dynamic approach of spearing its prey with the talons of one or both of its feet. This mode of manipulation, which clearly involves dynamic impact at the end e ector (here the talons), is one that is receiving increased attention by robotics researchers. This interest is due to the fact that many manipulation tasks can be completely transformed when approached from a dynamic contact (as opposed to quasistatic contact, the traditional approach in robotics) viewpoint. Many tasks (for example extracting a sh from water) can be performed much faster and more eciently with impulsive manipulation [6]. It has been previously noted that many animals, such as raccoons [13] appear to use dynamic manipulation modes extensively in nature, making good use of their relatively simple end e ectors.
Figure 7: Model of osprey foot used in dynamic analysis. sh from behind. As discussed earlier, this appears to be the case that the osprey follows in practice [8], and no doubt has the added advantage of allowing the bird to track the sh more accurately during descent. The outstretched con guration of the talons preferred by the osprey at the instant of impact minimizes the impact forces imparted to the toe joints.
To investigate the dynamic grasping behavior of the osprey, a dynamic simulation of its shing behavior was conducted. The osprey foot kinematics (see Figure 2) were approximated by the model in Figure 7 (where in both cases the shaded region denotes coupling between the displayed joints, and the joints connected by a shaded region represent one independent degree of freedom). Average mass and dimensional parameters for the osprey and sh were taken from [3] and [16]. For the moment of contact, the robot impact model in [15] was used. The sh was assumed to be a point mass, and the contact conditions to correspond to a plastic collision. The drag induced by the bird hitting the water was neglected for simplicity (in essence, the sh is assumed to be immediately below the surface of the water).
The impact force is also strongly a function of the normal to the plane of contact, which in this case corresponds to the geometry of the sh. The contact forces drop signi cantly (an order of magnitude) depending on whether the strike plane is parallel to the water surface (the highest value, corresponding to hitting the sh along its back), or more perpendicular to the water surface (much lower values corresponding to hitting the sh along the downward slope of its head or side). Interestingly, a combination of a dive angle of 45 degrees (the observed typical dive angle [8]) and a contact normal aligned with the dive angle result in an impact force from the model independent of the velocity of the bird! This is due to the fact that for rigid body collisions, the impulsive force is a function of the bodies' pre-collision velocity normal to the plane of contact [15]. In practice this implies that, by matching its dive angle to the plane of contact on the sh, the bird can achieve essentially the same impact (capturing) force on the sh over a wide range of velocities. Of course, there are practical constraints on the velocity of the bird due to the need to track the sh in the water before impact. However, this inherent possibility of achieving the same force task goal over a variety of closing contact velocities and con gurations with dynamic manipulation is intriguing when considered from the robotics point of view.
The results of the simulation con rm the e ectiveness of the osprey shing strategy. For the case of single talon contact (talons 1 and 3 were considered), the contact condition that resulted in the highest impact force imparted to the sh, (for the typical raptor closing velocity of 13:4 m=s) was when the raptor descended along the opposite direction to the sh trajectory. This is to be expected, and corresponds to a `head-on' collision, resulting in a high impulsive force imparted to the talons, which may be undesirable to the bird. Much smaller impact forces to the sh (of at least a factor of two) result from the opposite case, with the osprey following the sh trajectory, and hitting the
Thus, the results of the dynamic simulation of osprey 7
up to ve feet in length [8]. To bring these building supplies to the nest site, an osprey must y while holding the item in both feet, using only point contacts with friction. During landing, the bird releases one foot from its cargo and uses it to guide the twigs, through a series of impulses, into the correct position by using the foot still grasping as a pivot point. Once the nest's supporting structure has been built, a soft bed of materials suitable for eggs must be gathered. Osprey nests have been found which contain a large variety of trash comprising this layer. Grass, shing line, seaweed, and rope have all been reported [8], [10]. Because these materials are soft, we surmise they were brought to the nest by compliant grasping. This accounts for observations that an osprey may hold these materials in a single foot during ight. As we have seen, the formalism we have applied in examining the various contact situations seems to agree well with our intuition into everyday raptor grasping.
shing yield information which supports and provides new understanding of the osprey hunting behavior. In particular, the dynamic grasping strategy clearly compensates for the lack of kinematic dexterity in the hand. In addition, the shing example provides some insights into how dynamic manipulation with simple hands can be e ectively conducted. 4
Discussion
Having well established a perspective on raptorial contact situations and having examined a stereotypical dynamic grasping situation, we now consider the implications of our analyses on other raptor grasping tasks. Consider rst the task of eating. After having caught a sh, an osprey may choose to perch in a nearby tree to consume its prey. To do so, it must support its weight with one foot and hold the sh in the other. If the sh is fairly large compared to the size of the foot, the osprey will not pierce the sh with its talons, but will instead use a compliant grasp, encircling its food as much as possible to prevent it from getting away. While maintaining its balance on the other foot, it will lift the sh to its mouth and tear a bite-size piece o with its beak, e ectively introducing an impulse to the system which the grasp has to resist. This process is repeated until the remainder of the sh is about the size of the bird's foot. At this point, constraining the food with the whole foot encumbers the osprey's ability to consume the remaining meat. Thus, it must switch to a di erent contact regime. Our second contact case, tip piercing, takes over and the bird pierces its food with the tips of one or two talons. Constraining the food in this manner, our osprey consumes all but the tail and toughest of scales, allowing little to go to waste [10].
The kinematic analysis suggests that the osprey hand is quite limited. However, as we have seen, the osprey is able to display highly successful specialized grasping behaviors. Obviously, the human hand is signi cantly more versatile than the osprey foot. However, this is due to its signi cantly increased complexity, which we are not close to reproducing with robotic devices. The simpler design of the osprey foot presents a more achievable design with current technology. Thus, an interesting question that arises from the analysis in this paper is: are there aspects of the grasping behavior of the osprey which can be successfully used in the robotics arena? Several points arise in this regard. First, the use of the talons for piercing is clearly a key to overcoming the diculties in achieving force closure for the osprey. The use of `claws' on end e ectors is not an area that has received signi cant attention in robotics. Clearly, many applications preclude the use of talons or claws, due the the need to protect the environment or payload. However, for some applications this may represent a useful grasping mode.
When a male osprey is a member of a breeding pair, his chief responsibility is to provide food to his mate and young. Taking his share of the meal rst, the male osprey will consume the head of its prey as described above and deliver the rest to its mate and their young where the female will feed the young. In this context, the female is able to use only point contacts with friction (Case 4 above) to constrain the meal since it will o er her no resistance. She can achieve force closure by holding the sh against the oor of the nest. The osprey will tear o small pieces of meat with her beak and feed them to her hungry young. Other than the contact type employed, the rest of the process is the same as for general eating which was discussed earlier.
Secondly, the use of the environment to provide the `extra' ngers to achieve force closure is interesting. Almost all robots operate independently of the world as much as possible, attempting to achieve the entire task within the robot structure. The use of `props' in the world surrounding the robot is clearly a successful one for the osprey, as well as for many other creatures. Finally, the strong and successful use of dynamic manpulation by the osprey reveals how, by interacting dynamically with the world as opposed to the quasistatic mode almost universal in robotics, allows some tasks to be performed more eciently. In fact, in some cases, the task could not be performed at all without the dynamic manipulation mode. It appears that the issue of
Finally, let us consider the task of nest building. A young breeding pair may spend their entire rst season together constructing a nest. First, they build a frame for the nest in a (preferably dead) tree out of sticks and twigs. These rst components may be 8
dynamic manipulation exempli ed by the osprey can be a key mode in the future for robotics. 5
References [1] Corel Photo Studio. Image Number 591005 in Corel Photo CD Collection, 1997. [2] Cutkosky, M.R. Robotic Grasping and Fine Manipulation. Kluwer Academic Publishers, Boston, 1985. [3] Dunne, P. The Wind Masters: The Lives of North American Birds of Prey. Houghton Miin Company, Boston, 1995. [4] Gill, F.B. Ornithology. Second Edition. W.H. Freeman and Company, New York, 1995. [5] Grupen, R.A., Henderson, T.C. and I.D. McCammon. A Survey of General Purpose Manipulators. In The International Journal of Robotics Research, Vol. 8, No. 1, pages 38-62, 1989. [6] Huang, H.H., Krotkov, E.P., and M.T. Mason. Impulsive Manipulation. In Proceedings 1995 IEEE International Conference on Robotics and Automation, pages 120-125, Nagoya, Japan, 1995. [7] Lucas, G. The Empire Strikes Back. LucasFilm Ltd. Released by Twentieth Century Fox, 1980, 1997. [8] Male, M. and J. Fieth. Nova: Return of the Osprey. Blue Earth Films for WGBH Educational Foundation, 1986. [9] Mason, M.T. and J.K. Salisbury. Robot Hands and the Mechanics of Manipulation. MIT Press, Cambridge, Massachusetts, 1985. [10] Poole, A.F. Ospreys: Their Natural and Unnatural History. Cambridge University Press, Cambridge, United Kingdom, 1989. [11] Proctor, N.S. and P.J. Lynch. Manual of Ornithology: Avian Structure and Function. Yale University Press, New Haven, Conneticut, 1993. [12] Van Tyne, J. and A.J. Berger. Fundamentals of Ornithology. Second Edition. John Wiley & Sons, New York, 1976. [13] Walker, I.D. A Successful Multi ngered Hand Design|The Case of the Raccoon. In Proceedings 1995 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 186-193, Pittsburg, PA, 1995. [14] Walker, I.D. Modeling and Control of Mult ngered Hands: A Survey. To appear in Complex Robotic Systems, P. Chiacchio and S. Chiaverini, eds., Lecture Notes in Control and Information Sciences Series, Springer-Verlag, London, 1998. [15] Walker, I.D. Impact Con gurations and Measures for Kinematically Redundant and Multiple Armed Robot Systems. In IEEE Transactions on Robotics and Automation, Vol. 10, No. 5, pages 670-683, 1994. [16] Wheeler, B.K. and W.S. Clark. A Photographic Guide to North American Raptors. Academic Press, London, 1995.
Conclusions
In this paper, we have considered a highly successful example of a multi ngered end e ector in nature, namely the four-toed osprey foot. The osprey foot is of an unusual kinematic design, and the bird uses its feet in novel ways compared to the typical use of dextrous robotic end e ectors. We have used the concept of form and force closure to analyze the capability of the osprey foots kinematic design to achieve grasp stability. In addition, a dynamic analysis of the osprey in shing mode, in which the bird dives and dynamically captures the sh, has been conducted. The results of the analyses con rm several key physical factors underlying the grasping and manipulation strategy of the osprey. For quasistatic grasping situations, the bird's use of the environment to achieve grasp stability seems a necessary result of the relative simplicity of the foot kinematics. From the dynamic analysis one can see how the bird optimizes its ight trajectory and toe con guration to best achieve dynamic impact with the target when shing. From the perspective of robotics, the design of the osprey foot is appealing due to its relative simplicity and compactness. Robot hands based on the osprey design would be simpler and cheaper than those based on the human design. This could result in signi cant bene ts in robotic applications if the functionality of the hands was sucient for application tasks. A key issue here is the inherently dynamic way in which the osprey uses its feet in many manipulation tasks. The use of dynamic manipulation, which inherently implies dynamic impact, allows tasks to be performed faster and in many cases (for example osprey shing) more e ectively than using a quasistatic approach. It seems that many creatures with relatively simple multi ngered end e ectors, such as the osprey, successfully use dynamic manipulation to compensate for the lack of dexterity in their hands. This provides motivation for the robotics community to further investigate the issue of how the use of dynamic manipulation could allow many tasks to be undertaken in a more e ective manner, by simpler robot hand designs than those currently under investigation. This work was supported under NSF grant number CMS9796328, by the U.S. Department of Energy under contract number DE-FG07-97ER14830, and a grant from the Clemson University Research Grant Committee.
9
