Current Projects

The interaction of terrestrial biologic/robotic organisms with ground is fundamental to how such systems move. We hypothesize that basic interactions of a spring-loaded, inverted pendulum (SLIP) template with phenomenological ground models--including stiffness, damping, and coulomb friction cones will provide an explanation for several observed scaling principles and better insight into the process of tuning and control, which can be directly translated to robotic design and control.

Understanding the underlying dynamics of the Robotic Hexapod, RHex, is critical to predicting its performance, controlling its motion, and evolving its design through an iterative engineering process. We have developed an anchored model of the robot to determine which parts are essential to reproduce its dynamic behavior, and have found that key pieces of each subsystem must be included. Further, the model can exhibit chaos which suggests that a new strategy for predicting robot performance needs to be developed.

Back to top.

Morphology Effects in Locomotion

Justin Seipel

Movie: rigid_roach_3.mpg

To enable robotic locomotion over challenging terrain we need to understand how leg number and posture during stance affects the stability, maneuverability, robustness, and energetics of motion. We also wish to understand how control systems can interact with morphology and the environment. We can now quantify fundamental "first-order" effects from morphology and are currently developing more detailed, integrative models of the cockroach Blaberus discoidalis and the Robotic Hexapod, RHex. We have also developed fundamental models of back-bending in the horizontal and sagittal planes and are currently analyzing them to understand the principles of back-bending in locomotion and to design physical robotic models.

Back to top.

Gecko Adhesion

Anne Peattie

Movie: Single Seta Force.avi

We investigates the function and evolution of adhesive gecko feet. Geckos possess unique pads on their toes covered in tiny hairlike protrusions called "setae". Setae vary widely in morphology across geckos, from short spines to longer (150 microns), branched structures with 1-1000 flattened tips (called "spatulae"), each about 200nm wide. Setae are completely epidermally derived and are shed and regenerated along with the rest of the skin. They are primarily composed of beta-keratin.

Using phylogenetic comparative methods, our goal is to determine how morphological variation across species impacts adhesive performance. Do setae with more spatulae generate more adhesive force? Do longer setae perform better on rough surfaces than shorter setae? Is it better to pack many small setae densely, or fewer, bigger setae less densely? By combining biomechanical and phylogenetic techniques, we hope to reveal the processes driving gekkonid adhesive evolution.

Back to top.

Testing Neuromechanical Control Architectures

Obstacle Traversal

Shai Revzen

Movie: Obstacle Traversal run.avi

To study how running cockroaches recover from a simple, biologically plausible perturbation, we developed a system for running a cockroach in a cage on a transparent treadmill. A computer controlled obstacle can be moved across the floor of the cage at a controlled speed while the animal is filmed with high speed video from below. Software Shai Revzen developed allowed us to use a combination of automated and manually assisted tracking to obtain large volumes of data showing the position of the tarsi (feet) and the body. Revzen then applied a phase estimation technique he developed for this purpose to model the state-space trajectories of the animal and allow us to compare "before" and "after" running in novel ways. The results were very surprising and are now being prepared for publication. An initial version was presented in the SICB 2005 meeting.

Back to top.

Lateral Impulse

Shai Revzen

Jindrich and Full (2002) attached a jet-pack to the back of a cockroach and then fired sideways it as the animal was running and measured how long it took for the animal to recover from the impulse. To test hypotheses concerning neuromechanical feedback over a longer time period, we devised a new perturbation method. We know that "virtual" forces in non-inertial frames of reference are just as real as "real" forces. Therefore, we ran cockroaches onto a plate that can be rapidly accelerated in the lateral direction, and then continues inertially. In the plate frame of reference, the cockroaches experience a perfectly reproducible and homogeneous impulse.

Using an extension of the software tools Shai Revzen developed for the obstacle perturbation experiment, he tracks the plate motion and compensate for it and the perspective changes it induces. The motion compensated video is tracked with the cockroach tarsus tracking tools to provide a comparable dataset to that obtained in the obstacle experiment. Analysis of the state-space model and its changes is underway. The results support and extend Jindrich's discovery and demonstrate how the mechanisms predicted in the analysis of the obstacle perturbations come into play.

A presentation of these results was given in the special symposium honoring Steven Vogel in the SICB 2006 meeting.

Back to top.

Antennal Tactile Navigation

Simon Sponberg

Movies: Antenna wall-following.mov Antenna wall-following threeobsticles.mov

Understanding the neuromechanical control of stability is necessary, but animals also use information from their environment to navigate. In addition to perturbation rejection understanding control of locomotion requires investigating task level challenges where animals must make alterations to their steady-state locomotion to maneuver. Besides being remarkably fast runners (up to 1.5 m/s) American cockroaches (Periplaneta americana) demonstrate remarkable tactile navigation ability. P. americana can track surfaces with antennae held out in front of the body during running. They can successfully execute up to 25 tactile stimulated turns/second while tracking an accordion shaped wall and can begin to respond to a change in wall angle in as little as 30 ms. Besides providing a tractable experimental system, amenable to modeling, behavioral analyses, and neurophysiological techniques, such extraordinary nature performance suggests that tactile sensors may provide a valuable control strategy for robots operating with limited visual salience. In collaboration with Professor Noah Cowan's lab at John Hopkins University, we are seeking to model and test control hypotheses for this behavior in both biological and physical systems.

Simple biomechanical control models have shown that the cockroach's dynamic response to tracking a step change in wall angle is unstable if proportional information (distance from wall) alone is used. The model predicts that the rate of change of distance from the wall (derivative information) is required from sensors. The simplest source of derivative control predicts a phasic response in antennal mechanoreceptors during an increase in bending - equivalent to a change in wall orientation. We recorded extracellularly from the antennal nerve of restrained American cockroaches using sharpened tungsten electrodes. The antenna was flexed both with a point deflection and using a mechanically actuated wall segment to simulate turns during running. We calculated the rms value of the net antenna nerve response to describe the power in the neural signal. Neural activity significantly increased during and immediately after the stimulus followed by a subsequent decrease from peak power. This phasic response was followed by a long-time constant "tonic" response when the antenna was held in a deflected position. If the antenna was deflected towards the cockroach, this response showed an increase over baseline neural power. If the deflection was away from the body, the signal's power decreased. Our results provide the first neurophysiological tests supporting a more complex (non-strictly proportional) controller for antenna wall-following, and suggest that the necessary information for control may already be present in the nervous system immediately downstream from the antennal mechanoreceptors.

The results for this first experiment are consistent with the hypothesized proportional derivative control model and the time course of the neural signal matches closely the timing observed in kinematic trials of a roach following a turn in a wall. However, to characterize the actual biological controller we must record the activity of individual neurons, extracted time constants of the neural response, and explicitly defined the position-, velocity- and frequency- response of antennal mechanoreceptors. The typical analysis called for in this situation is a white noise stimulus experiment over many repeated trials. This typically requires single unit resolution to determine response functions and peak time constants. In collaboration with Professor John Miller's lab at the Center for Computational Biology at Montana State University we are developing a new en passant suction electrode recording method for both bulk and individual unit recordings of the antennal nerve in a restrained prep. This new technique should enable a thorough testing of alternative complex control hypotheses and the design of a biologically-inspired controller for tactile navigation in robots.

Back to top.

Rough Terrain

Simon Sponberg

Stability is critical in animal locomotion. Speed, efficiency, and economy might matter little if common perturbations were sufficient to destabilize a running, flying, or swimming animal. Many studies have shown that neural feedback, often in the form of reflexes, contributes to the maintenance of stability in locomoting animals. Recently models of the physical properties and dynamics of legged runners have shown that stability could also result from mechanical feedback, the passive physical response of the system to perturbations. These control strategies, neural and mechanical feedback, can, and likely do, both contribute to control in many behavioral contexts. They define a neuromechanical control space in which the organism operates. We are currently investigating where in this control space organisms operate during typical behaviors.

Discoid cockroaches are able to successfully navigate complex terrain with obstacles up to three times their hip height without loosing stability. From previous studies it is known that a suite of femoral extensor muscles in the cockroach's hind leg are important for responding to vertical obstacles during locomotion. To investigate where in the neuromechanical control space a cockroach operates during preferred speed locomotion, we recorded the neural activation of these muscles as a cockroach runs over a random rough surface. When neural feedback acts to stabilize an animal after it is perturbed, there are changes in the activation patterns stimulating the organism's muscles. Mechanical feedback does not require changes in muscle activation patterns. By investigating the extent muscle activation patterns change, we can identify where cockroach's control strategy lies along the axis of neural and mechanical feedback.

Earlier studies have demonstrated that during slow exploration the control of these muscles seems to reside more towards the neural feedback region of the control spectrum. Surprisingly, in our experiments we did not detect changes in these muscles while traversing the rough terrain, suggesting that cockroaches make extensive use of mechanical feedback during preferred speed locomotion. Importantly, when we challenged the cockroach with even greater obstacles, up to 6 times their hip height we observed regular changes in the neural activation patterns indicating that neural feedback was acting on these muscles. Together these results suggest a hypothesis of nested stability basins where animals locomoting at high-speed with significant bandwidth constraints on control rely more on mechanical feedback to reject small to moderate perturbations, while using a larger region of stability from neural feedback to maneuver and react to large perturbations.

Read More
2005. Scurrying Roaches Outwit Without Their Brains. Science, vol. 307, p. 346. Full text of this article in PDF

Back to top.

Rewriting Motor Activation Patterns

Simon Sponberg

Movies: Spike Addition1_divx.avi Spike Addition2_divx.avi

While extensive research has uncovered the sensory (afferent) encoding of many different modalities (light, sound, odor) in a diverse set of organisms, a parallel understanding of how neural output is encoded in the motor (efferent) side of the motor system is lacking. In fact, while the input S(t) for a sensory system can often be well described (e.g. the power spectrum of an acoustic stimulus), and we can capture the time course of the resulting neural spike trains R(t), the corresponding output, K(t), encompassing the behavioral output of a motor system is ill-defined. Additionally, we know that the physical dynamics of a locomoting organism alone can lead to stability, yet reflex feedback control is clearly important in many tasks. This dual problem of unraveling motor code output and determining the relative importances of reflex control in different behaviors is critical for developing generative models of the control strategies used by animals that could serve as the basis for the next generation of legged robots.

Using biomechanical models to mathematically describe locomotion (e.g. as a reduced set of principle components of joint kinematics or limit cycles of center-of-mass kinetics), we can now relate changes in the neural-muscular code to changes in the mechanical output of the system. Experimentally, we have developed a system to begin rewriting the motor activation patterns of a behaving organism by simultaneously recording muscle activation during running, modifying this activation by stimulation at specific phases during locomotion, tracking the movement of the cockroach's center of mass and leg trajectories using high-speed video, and measuring the accelerations affecting the center of mass using a 3-dimensional micro-accelerometer backpack. In the first experiments using these methods to test a specific suite of control muscles in the cockroach, Blaberus discoidalis, we have demonstrated that a muscle's potential for control and the information its activation pattern conveys to the mechanical system is behaviorally dependent. Specifically, the same changes in muscle activation that can produce graded forces on the center of mass while the animal is stationary have only very gross control over the center of mass dynamics during high-speed locomotion. This challenges the canonical interpretation of past reflex studies where active changes in the neural activation patterns to muscles in response to perturbations were characterized as static rules for locomotion. More broadly, these results point to the underlying theme that to understand locomotion, we must investigate how active, reflex-mediated control interacts with the tuned musculo-skeletal system of locomoting animals to generate stable, versatile behaviors.

Back to top.

Collaborative Projects

Using Bio-inspired Robots as Physical Models in the Teaching Laboratory

Joel Weingarten, Haldun Komsuoglu, Tim Kubow Professor Daniel Koditschek and Professor Robert Full, Univ. of Pennsylvania and Univ. of California, Berkeley

We have developed a bio-inspired robot named Edubot that can be used in both the teaching and research laboratory. Edubot is a programmable, hexapedal robot patterned after the robot, RHex, that can demonstrate the dynamics discovered in legged animals. Edubot can serve as a physical model to reveal critical parameters missing from mathematical models, because like an animal the robot must confront the real environment. In a new, introductory undergraduate engineering course at the University of Pennsylvania, research-based teaching laboratories that complement a novel freshman introduction to Electrical and Systems Engineering have been developed and implemented with open-ended problems that include programming the robot to traverse rough terrain autonomously using only accelerometers, designing new legs to handle obstacles and creating a novel robot dance involving two thousand leg movements that demands the development of a hierarchical controller integrated with the leg and body dynamics. In collaboration, we are presently designing a complementary research-based course at the University of California at Berkeley in the Department of Integrative Biology as a critical part of a new center Ð CIBER Ð the Center for Interdisciplinary Bio-inspiration in Education and Research. We aim to eliminate the traditional (but misconceived) gaps between research and teaching. It is time for a synergy where research-based teaching leads to better learning and collaborative, interdisciplinary teaching leads to better research. Funded by NSF FIBR grant.

Back to top.

Mechanics

Many-legged, Sprawled Posture Stability

• Stability of hexapods to rapid impulse perturbations - Devin Jindrich
• Hypothesis generation using dynamic models - Dr. Mariano Garcia, Tim Kubow, Anna Ahn, and Devin Jindrich
• Exo-skeletal joint damping in many-jointed animals - Dr. Mariano Garcia, Dr. Kenneth Meijer, Anne Peattie, and Paul Wang
• Insect legs as springs - Dan Dudek
• Posture principles from reducing DOF using PCA - Dr. Mariano Garcia
• Joint power in hexapods - Dr. Mariano Garcia
• Learning many-legged sprawled posture locomotion - Dr. Mariano Garcia
• Differential leg function of geckos running on the level - Juliann Chen
• Mechanics of centipede locomotion - Bruce Anderson

Many-legged, Sprawled Posture Maneuverability

• Dynamics of vertical climbing in geckos - Prof. Kellar Autumn
• Adhesive force of a single gecko foot hair - Prof. Kellar Autumn, Tonia Hsieh, and W. Pang Chan
• Dynamics of turning in hexapods - Devin Jindrich
• Dynamics of insects climbing a single vertical step - Prof. Kellar Autumn, Dr. James Glasheen, Raul Trejo, and Amy Jagger

MUSCLE

The Role of Muscle in Locomotion and Control

• Performance of preflexes during locomotion - Dr. Kenneth Meijer
• Differential function for muscles of an anatomical group - Anna Ahn

EXERCISE

Intermittent Locomotion

• Intermittent locomotion in geckos - Prof. Randi Weinstein
• Intermittent locomotion in highly aerobic insects - Prof. Randi Weinstein
• Intermittent locomotion in highly anerobic crabs - Prof. Randi Weinstein

ENERGETICS

The Cost of Maneuverability

• Energetic cost of insect on surfaces that differ in compliance - K. Chang
• Energetics of obstacle negotiate - I. Tan


Back to top.