Undergraduate Category: Interdisciplinary Topics, Centers and ...
Undergraduate Category: Interdisciplinary Topics, Centers and Institutes Degree Seeking: B.S. in Behavioral Neuroscience Abstract ID# 1581
Robotic Lobster Antennae Mimic Exploratory Sensory Behavior Jaimie Spahr and Joseph Ayers Opportunity Abstract In Dr. Ayers' lab, biomimetic robots are used as embodied simulations to analyze and reproduce marine invertebrate motor behavior. These submersible robots are meant to locate underwater mines autonomously using electronic neurons in a discrete-time map-based (DTM) network. The robotic lobster is mobile underwater, detects agents of harm, and can maneuver around water currents and surges. Natural lobsters have large antennae that are approximately the same length as the distance they can reach, so they use their antennae for active exploration to determine if it may interact with the object. Building upon the existing capabilities of the robots, this project will develop and implement code to replicate antennal behavior. This includes the construction of the antennae, measurement of antennal response to its environment, and the consequential behaviors triggered by this sensory information. The project allows RoboLobster to interact with its environment and maneuver around obstacles in its search for underwater mines.
Approach
Data & Results
The internal electronics of the robots are based on neuronal circuits, which communicate with the mechanical body and sensors just as neurons in the brain communicate with afferent and efferent nerves in the peripheral nervous system (Ayers, et al., 2010). These electronic neurons employ nonlinear dynamics to model network behavior instead of physiology in order to allow processing and action in real time. The neuronal networks are coded using neuron and synapse objects previously developed in LabView software and finally implemented in procedural C on Arduino board hardware, along with a variety of custom mechanical pieces.
When the antennae encounters an object in its environment, it is deflected or bent more than its baseline due to flow or its own movement. Deflection can be measured in a few ways, but for this project, sensors with variable resistance that changes as a result of deflection were used.
On the right are the strain gauge (top) and FlexPoint sensor (below). Both have an x-axis of time in milliseconds and a y-axis of a scalar for voltage. The values here are arbitrary, but the graphs provide visual evidence of the FlexPoint sensor’s preferred clarity, control, and sensitivity.
Introduction Invertebrates are used to study neural circuitry since they have relatively small numbers of neurons when compared to mammalian models. Lobsters are especially good for such studies due to the capability to reproducibly identify neurons (Ayers, et al, 2013, Matheson 2002). That is, much of circuitry is well documented and the neurons themselves are quite large, so it is easier to locate and therefore identify the same neuron across individual lobsters. Dr. Ayers was initially approached by government officials to take control of a lobster’s nervous system so that a person could navigate it as a vehicle to find underwater threats. Instead, he offered to build a robot which would run on electronic neurons in order to simulate living lobsters.
The natural lobster has two large antennae that it uses to feel around its environment and two pairs of smaller antennules for sensing odors (Factor 1995). They are able to gather information about the lobster's surroundings both passively and actively. The active exploration with the large antennae begins with visual input; something in the field of view catches the lobster's attention. Since the large antennae are about the same length as the distance the lobster can reach, they are waved back and forth to determine if it can grab the stimulus. This antennae behavior is thoroughly documented, so it lends well to robotic implementation. Building upon the existing capabilities of the robots in Dr. Ayers’ lab, this project explores the lobster’s neural network functionality and implement code to replicate antennae behavior.
Aim The goal of this work is to produce functional active touch antennae exploration in the robotic lobster.
First, the proper materials for the antennae were gathered and tested. Strain gauges and FlexPoint Bend Sensors of different lengths were compared. Arduino’s IDE was used as it is ideal for quick modifications during the prototyping process. The FlexPoint Bend Sensors were larger, more durable, and simpler to incorporate into a circuit than the strain gauges. The strain gauge being minimal hardware, it would require a wheatstone bridge to read the voltage, where the bend sensor is made to output the voltage with only a simple voltage divider set up. The 2” bidirectional sensor was determined to be the most sensitive of the FlexPoint sensors.
The image on the right is a diagram of the neuron network for the robot’s antennae. Rheotaxis refers to the movement of aquatic animals in response to currents. Active touch is where the lobster waves its antennae to explore the environment. Here, the rheotaxic interneurons stimulate the abductor neuron. The abductor neuron activates the muscles to spread the antennae apart, and the adductor neuron brings them together medially. The adductor neuron will be implemented as an endogenous burster, which means it produces bursts of action potentials when excited for prolonged periods of time. There is an inhibitory relationship between the abductor neuron and the adductor neuron, since they have opposing activity.
The FlexPoint sensor, when set up with a voltage divider, can provide the change in voltage or resistance to the Arduino board. The conductive ink particles printed on the sensor spread apart when bent one direction and squeeze together when bent the other, changing the conductivity of the sensor and therefore the resistance. The sensor is then caulked onto a cut polycarbonate shape that gives the following prototype:
Currently, the sensors used in the project work both on their own and attached to the polycarbonate antennae. A simple voltage divider and an Arduino Uno are used in the circuit to read voltages from the sensor. I am developing code to calibrate the sensors for initial testing, which includes defining the sensor’s sensitivity in an equation that will be used for environment prediction. Future steps broadly include: • Calibration and definition while underwater • Develop LabView code for antennae • Test in LegoRobot networks • Incorporate into the existing RoboLobster neural network
Impact Discussion The antennae ultimately give the RoboLobster sensation environmental obstacles that would otherwise go undetected.
of
Electronic neurons make this approach unique, because the majority of autonomous robots rely on traditional artificial intelligence methods. The completion of this project not only more accurately represents a living lobster's behavior, but gives greater abilities to the existing robots that are used for detection of agents of harm underwater. In addition to chemical detection, the tactile behaviors will allow for physical detection and downstream processes that lend well to other applications. Replicating more neural activity accurately in a computer, in this case a robotic lobster, represents an increase in the field’s understanding of all neural networks and how the basic motor movements function even in the human brain. The network becomes an embodied hypothesis of how the animal controls active touch. While this project focuses on a single behavior of just one species, this work, and the other research going on in the lab, sets the foundation for more complex robots and humancomputer interfaces that many tech companies will be investigating in the coming years.
References Ayers, J. (2004). "Underwater walking." Arthropod Structure & Development. 33(3): 347-360. Ayers, J., Rulkov, N., Knudsen, D., Kim, Y-B., Volkovskii, A., Selverston, A. (2010). Controlling Underwater Robots with Electronic Nervous Systems. Applied Bionics and Biomimetics. 7: 57-67. Ayers, J., et al. (2012). A Conserved Biomimetic Control Architecture for Walking, Swimming and Flying Robots. Biomimetic and Biohybrid Systems. 7375: 1-12. Biomimetic Underwater Robot Program. (2010). Northeastern Marine Science Center. Retrieved from: www.neurotechnology.neu.edu/ Factor, J.R. (1995). Biology of the lobster Homarus americanus. San Diego, Academic Press. Macmillan, D. L., et al. (1980). Neurobiology. The Biology and Management of Lobsters: Physiology and Behavior. Vol.1: 165214. Matheson, T. (2002). Invertebrate Nervous Systems. Encyclopedia of Life Sciences. Retrieved from: https://www2.le.ac.uk/departments/ npb/people/matheson/matheson-neurobiology/images/ publications/Matheson_ELS_2002.pdf Wilkens L. A., B. Schmitz and F. Herrnkind (1996). "Antennal responses to hydrodynamic and tactile stimuli in the spiny lobster Panulirus argus." Biological Bulletin (Woods Hole) 191(2): 187-198. Zeil, J., R. Sandeman and D. Sandeman (1985). “Tactile localisation: the function of active antennal movements in the crayfish Cherax destructor." Journal of Comparative Physiology A: Sensory, Neural, and Behavioral Physiology (Historical Archive) 157(5): 607.
Acknowledgements Thank you to the Office of Naval Research for supporting this research (ONR MURI N000141110725) and to Dr. Ayers for providing me mentorship and the opportunity to work in his lab. A special thanks to Ryan Meyers for his advice and help in the lab over the year.
Robotic Lobster Antennae Mimic Exploratory Sensory Behavior Jaimie Spahr and Joseph Ayers Opportunity Abstract In Dr. Ayers' lab, biomimetic robots are used as embodied simulations to analyze and reproduce marine invertebrate motor behavior. These submersible robots are meant to locate underwater mines autonomously using electronic neurons in a discrete-time map-based (DTM) network. The robotic lobster is mobile underwater, detects agents of harm, and can maneuver around water currents and surges. Natural lobsters have large antennae that are approximately the same length as the distance they can reach, so they use their antennae for active exploration to determine if it may interact with the object. Building upon the existing capabilities of the robots, this project will develop and implement code to replicate antennal behavior. This includes the construction of the antennae, measurement of antennal response to its environment, and the consequential behaviors triggered by this sensory information. The project allows RoboLobster to interact with its environment and maneuver around obstacles in its search for underwater mines.
Approach
Data & Results
The internal electronics of the robots are based on neuronal circuits, which communicate with the mechanical body and sensors just as neurons in the brain communicate with afferent and efferent nerves in the peripheral nervous system (Ayers, et al., 2010). These electronic neurons employ nonlinear dynamics to model network behavior instead of physiology in order to allow processing and action in real time. The neuronal networks are coded using neuron and synapse objects previously developed in LabView software and finally implemented in procedural C on Arduino board hardware, along with a variety of custom mechanical pieces.
When the antennae encounters an object in its environment, it is deflected or bent more than its baseline due to flow or its own movement. Deflection can be measured in a few ways, but for this project, sensors with variable resistance that changes as a result of deflection were used.
On the right are the strain gauge (top) and FlexPoint sensor (below). Both have an x-axis of time in milliseconds and a y-axis of a scalar for voltage. The values here are arbitrary, but the graphs provide visual evidence of the FlexPoint sensor’s preferred clarity, control, and sensitivity.
Introduction Invertebrates are used to study neural circuitry since they have relatively small numbers of neurons when compared to mammalian models. Lobsters are especially good for such studies due to the capability to reproducibly identify neurons (Ayers, et al, 2013, Matheson 2002). That is, much of circuitry is well documented and the neurons themselves are quite large, so it is easier to locate and therefore identify the same neuron across individual lobsters. Dr. Ayers was initially approached by government officials to take control of a lobster’s nervous system so that a person could navigate it as a vehicle to find underwater threats. Instead, he offered to build a robot which would run on electronic neurons in order to simulate living lobsters.
The natural lobster has two large antennae that it uses to feel around its environment and two pairs of smaller antennules for sensing odors (Factor 1995). They are able to gather information about the lobster's surroundings both passively and actively. The active exploration with the large antennae begins with visual input; something in the field of view catches the lobster's attention. Since the large antennae are about the same length as the distance the lobster can reach, they are waved back and forth to determine if it can grab the stimulus. This antennae behavior is thoroughly documented, so it lends well to robotic implementation. Building upon the existing capabilities of the robots in Dr. Ayers’ lab, this project explores the lobster’s neural network functionality and implement code to replicate antennae behavior.
Aim The goal of this work is to produce functional active touch antennae exploration in the robotic lobster.
First, the proper materials for the antennae were gathered and tested. Strain gauges and FlexPoint Bend Sensors of different lengths were compared. Arduino’s IDE was used as it is ideal for quick modifications during the prototyping process. The FlexPoint Bend Sensors were larger, more durable, and simpler to incorporate into a circuit than the strain gauges. The strain gauge being minimal hardware, it would require a wheatstone bridge to read the voltage, where the bend sensor is made to output the voltage with only a simple voltage divider set up. The 2” bidirectional sensor was determined to be the most sensitive of the FlexPoint sensors.
The image on the right is a diagram of the neuron network for the robot’s antennae. Rheotaxis refers to the movement of aquatic animals in response to currents. Active touch is where the lobster waves its antennae to explore the environment. Here, the rheotaxic interneurons stimulate the abductor neuron. The abductor neuron activates the muscles to spread the antennae apart, and the adductor neuron brings them together medially. The adductor neuron will be implemented as an endogenous burster, which means it produces bursts of action potentials when excited for prolonged periods of time. There is an inhibitory relationship between the abductor neuron and the adductor neuron, since they have opposing activity.
The FlexPoint sensor, when set up with a voltage divider, can provide the change in voltage or resistance to the Arduino board. The conductive ink particles printed on the sensor spread apart when bent one direction and squeeze together when bent the other, changing the conductivity of the sensor and therefore the resistance. The sensor is then caulked onto a cut polycarbonate shape that gives the following prototype:
Currently, the sensors used in the project work both on their own and attached to the polycarbonate antennae. A simple voltage divider and an Arduino Uno are used in the circuit to read voltages from the sensor. I am developing code to calibrate the sensors for initial testing, which includes defining the sensor’s sensitivity in an equation that will be used for environment prediction. Future steps broadly include: • Calibration and definition while underwater • Develop LabView code for antennae • Test in LegoRobot networks • Incorporate into the existing RoboLobster neural network
Impact Discussion The antennae ultimately give the RoboLobster sensation environmental obstacles that would otherwise go undetected.
of
Electronic neurons make this approach unique, because the majority of autonomous robots rely on traditional artificial intelligence methods. The completion of this project not only more accurately represents a living lobster's behavior, but gives greater abilities to the existing robots that are used for detection of agents of harm underwater. In addition to chemical detection, the tactile behaviors will allow for physical detection and downstream processes that lend well to other applications. Replicating more neural activity accurately in a computer, in this case a robotic lobster, represents an increase in the field’s understanding of all neural networks and how the basic motor movements function even in the human brain. The network becomes an embodied hypothesis of how the animal controls active touch. While this project focuses on a single behavior of just one species, this work, and the other research going on in the lab, sets the foundation for more complex robots and humancomputer interfaces that many tech companies will be investigating in the coming years.
References Ayers, J. (2004). "Underwater walking." Arthropod Structure & Development. 33(3): 347-360. Ayers, J., Rulkov, N., Knudsen, D., Kim, Y-B., Volkovskii, A., Selverston, A. (2010). Controlling Underwater Robots with Electronic Nervous Systems. Applied Bionics and Biomimetics. 7: 57-67. Ayers, J., et al. (2012). A Conserved Biomimetic Control Architecture for Walking, Swimming and Flying Robots. Biomimetic and Biohybrid Systems. 7375: 1-12. Biomimetic Underwater Robot Program. (2010). Northeastern Marine Science Center. Retrieved from: www.neurotechnology.neu.edu/ Factor, J.R. (1995). Biology of the lobster Homarus americanus. San Diego, Academic Press. Macmillan, D. L., et al. (1980). Neurobiology. The Biology and Management of Lobsters: Physiology and Behavior. Vol.1: 165214. Matheson, T. (2002). Invertebrate Nervous Systems. Encyclopedia of Life Sciences. Retrieved from: https://www2.le.ac.uk/departments/ npb/people/matheson/matheson-neurobiology/images/ publications/Matheson_ELS_2002.pdf Wilkens L. A., B. Schmitz and F. Herrnkind (1996). "Antennal responses to hydrodynamic and tactile stimuli in the spiny lobster Panulirus argus." Biological Bulletin (Woods Hole) 191(2): 187-198. Zeil, J., R. Sandeman and D. Sandeman (1985). “Tactile localisation: the function of active antennal movements in the crayfish Cherax destructor." Journal of Comparative Physiology A: Sensory, Neural, and Behavioral Physiology (Historical Archive) 157(5): 607.
Acknowledgements Thank you to the Office of Naval Research for supporting this research (ONR MURI N000141110725) and to Dr. Ayers for providing me mentorship and the opportunity to work in his lab. A special thanks to Ryan Meyers for his advice and help in the lab over the year.
