Bulletin of the American Physical Society
APS March Meeting 2021
Volume 66, Number 1
Monday–Friday, March 15–19, 2021; Virtual; Time Zone: Central Daylight Time, USA
Session F14: Robophysics: Robotics Meets Physics II: Legged locomotionFocus Session Live Undergrad Friendly
|
Hide Abstracts |
Sponsoring Units: DBIO DSOFT Chair: Chen Li, Johns Hopkins University; Daniel Goldman, Georgia Inst of Tech |
Tuesday, March 16, 2021 11:30AM - 11:42AM Live |
F14.00001: Calibration Challenges for Measuring Multi-Legged Ground Contact Forces Ziyou Wu, Shai Revzen Ground contact modeling for multi-legged locomotion is challenging, because of the coupling effect of legs and foot slippage. We have previously presented a non-holonomic multi-legged locomotion model that used a mixed Coulomb-viscous-friction ansatz. This model matched experimental motion tracking data of multiple robots. To validate its predictions for the contact forces, we integrated a robot with six high precision 6DOF force-torque sensors. We present the force-torque sensing framework used in our robot and the present challenging calibration problems that arise. We hope that once calibrated, we will be able to directly and simultaneously measure the contact wrenches of all six legs. |
Tuesday, March 16, 2021 11:42AM - 11:54AM Live |
F14.00002: Leveraging Aerodynamic Drag for Tails in Legged Robot Locomotion Joseph Norby, Jun Yang Li, Cameron Selby, Amir Patel, Aaron Johnson The interaction between legged robots and their environment has largely been prescribed by the manipulation of contact forces. Yet nature has shown that other physical phenomenon can aid locomotion, such as swinging a tail to maintain balance or accelerate. Many biological tails are lightweight and furry unlike their robotic counterparts, suggesting a different orientation regulation mechanism than inertial effects. Motivated by this insight we explore the physics and utility of aerodynamic drag in tailed reorientation tasks. We present a model of the dynamics induced by aerodynamic drag and from this derive a metric that allows for direct comparison between aerodynamic and inertial tails. We draw several design and control insights and demonstrate their utility in two dynamic behaviors executed on a legged robot. We show that employing aerodynamic drag alongside inertial effects allows for a 63% reduction in normalized tail inertia when executing an aerial reorientation task. We also show a 12% increase in forward acceleration of the robot despite a 10% increase in system mass from equipping the tail. These results show that aerodynamic drag has the ability to transform the way robots interact with the world by expanding control authority independent of contact. |
Tuesday, March 16, 2021 11:54AM - 12:06PM Live |
F14.00003: Towards an Insect Inspired Segmented Robot Leg for Passive Deflection Over Steps Emily Lathrop, Michael T Tolley, Nick Gravish When observing insects walking over uneven terrain, we often see legs colliding with ground obstacles but quickly recovering by deflecting over these obstacles. Previous work has observed this behavior in both cockroaches and ants in both laboratory and natural settings. Given the rapid speed of locomotion and observations of quick leg recovery, it is hypothesized that foot-object interactions may be primarily aided by passive mechanical properties rather than through reactive control. These observations from nature have inspired our research into legged robotic solutions that can mimic insects by passively deflecting over obstacles encountered in the swing phase. Inspired by the multi-segmented architecture of insect legs, we have developed a 3D printed segmented robot leg that allows a walking robot to passively deflect over obstacles in the swing phase and bear weight during the stance phase. We performed walking tests over laboratory models of uneven terrain such as vertical steps and we observe that segmented multi-stiffness legs aid in successful walking. This result offers new design choices for dynamically stable walking robots over natural terrain. |
Tuesday, March 16, 2021 12:06PM - 12:18PM Live |
F14.00004: Arachnids as systems for robophysical study Joseph Spagna Arthropods in the class Arachnida (inclusive of 8 orders and over 100,000 species) demonstrate suites of remarkable adaptive biological features, which lend themselves to robophysical study. The arachnid body plan is two body sections and four pairs of jointed walking legs extending from the anterior body section. In addition, they have a fifth set of leg-homologs in front of the legs that serve as sensors, grabbers, or sex organs. The running mechanics and leg features of spiders have been modeled and tested using the rHEX and RiSE robotic platforms, respectively, but a range of opportunities for study remain. Arachnids, with eight legs and a tendency to autotomize them when in danger- make an ideal study system for leg redundancy and flexibility in limb usage after loss. Solifugae (“wind-scorpions”) have adapted to running on flowing substrates at steep angles; the mechanism for this is unknown. Finally, spiders make prey traps (webs) using small, variously spaced protein strands. Each of these systems suggests hypotheses of selection leading to optimization, but empirical data are ambiguous. We suggest that for these varied arachnid adaptations, robotic systems can be used to better test- and possibly optimize- key mechanisms of these systems and their analogs. |
Tuesday, March 16, 2021 12:18PM - 12:30PM Live |
F14.00005: A Motion Planning Framework for Dynamic Monopedal Locomotion on Deformable Terrain Daniel Lynch, Kevin M Lynch, Paul Byron Umbanhowar The state-of-the-art in robotic legged locomotion on deformable terrain remains many steps behind the stable, graceful, and agile gaits exhibited by many animals that walk, trot, and run over soft substrates like sand, soil, and snow. A defining trait of legged locomotion on these deformable surfaces is simultaneous intrusion and rotation of the foot, which is neglected in most robotic locomotion planning and control frameworks such as zero moment point control and hybrid zero dynamics. We address this gap, and extend the work of Li et al. (2013) and Xiong et al. (2017), by developing a motion planning framework that fuses resistive force theory with direct collocation trajectory optimization to plan foot trajectories through granular media. We apply our framework to the task of finding period-1 hopping gaits for a monopedal robot. We first formulate a collection of constraints that represent period-1 orbits in state space, and we then use our framework to obtain motions that satisfy these constraints, i.e., monopedal hopping gaits. We find that these gaits scale meaningfully with speed and ground stiffness, suggesting that our approach has the potential to close the gap between robotic and biological legged locomotors. |
Tuesday, March 16, 2021 12:30PM - 12:42PM Live |
F14.00006: Optimizing Contact Area and Joint Stiffness of a Passive Foot-Ankle Exoskeleton for Locomotion on Deformable Terrain Jonathan Gosyne, Gregory S. Sawicki Designing exoskeletons to improve locomotion energetics on deformable terrain is an engineering challenge. This is due to complex, dissipative boundary conditions at the foot-ground interface that cause unsteady muscle tendon (MT) dynamics. During ground contact, the MT is required to perform net positive work to overcome energy loss to the environment. This creates an opportunity for assistive technology that targets muscle level function to improve economy/efficiency of force/work generation in dissipative environments. As a first step, we developed a musculoskeletal model to understand device-MT interaction in order to predict changes in metabolic energy cost (Pmet) on solid ground vs. sand. We modelled an exoskeletal device with variable ankle stiffness and foot contact area coupled to a Hill-Type model representing the calf-Achilles tendon MT. During fixed frequency locomotion cycles, we found terrain-specific exoskeleton parameters that eliminated the Pmet penalty incurred during locomotion on sand vs. solid ground. This preliminary work in-silico to optimize exoskeleton parameters offline will drive the development of novel hardware solutions to augment performance at the human-machine interface in unstructured, real world environments. |
Tuesday, March 16, 2021 12:42PM - 12:54PM Live |
F14.00007: An insect-scale robot reveals the effects of different body dynamics regimes during open-loop running in feature-laden terrain. PERRIN SCHIEBEL, Henry Cerbone, Robert J Wood Insects and robots can traverse heterogeneous terrains using open-loop dynamics. However, robots typically operate within a narrow range of running frequencies dictated by body and leg resonances; the impact of different dynamics regimes is largely unknown. We used the quadrupedal Harvard Ambulatory MicroRobot to study trot and pronk gaits at vertical and pitch/roll body resonances and above resonance. We tested open-loop running on a homogeneous substrate and two 3D printed heterogeneous terrains with foot-size features. The "pebbly" terrain features were shorter than limb amplitude while the "alpine" terrain features increased linearly to above limb amplitude. Distance traveled per gait cycle on the homogeneous and pebbly terrains were linearly related, suggesting above-resonance running which performs well in laboratory settings will also be successful on natural surfaces, provided the features are low enough to step over. In the alpine terrain, running in the body resonance regime bypassed the tallest obstacles. These results indicate that with rudimentary knowledge of terrain feature sizes, locomotors can target different body dynamics for effective open-loop running. This strategy may be useful for animals and robots with limited sensing and computing capabilities. |
Tuesday, March 16, 2021 12:54PM - 1:06PM Live |
F14.00008: Reactive velocity control reduces the energetic cost of transport for a direct-drive robot on granular media Sonia Roberts, Daniel E. Koditschek The Minitaur robot has direct-drive legs which can be used as force sensors to detect ground properties of interest to geoscientists studying erosion in deserts. However, Minitaur overheats quickly when running on sand. We used a combination of simulations, physical emulations with a ground emulator robot, and experiments on granular media to show that the energetic cost of transport for direct-drive robots on highly dissipative substrates like sand can be consistently reduced by at least 20% with no loss to jump height by adding a virtual damping force to slow the intrusion of the robot’s foot. The simulations used a bulk-behavior model of the forces exerted by granular media in response to intrusion. In the emulations, a single leg jumped on a highly transparent low gear-ratio ground robot leg programmed to exert a simplified version of the bulk-behavior granular media force model using PD control. The depth-dependent force determined the proportional gain and the velocity-dependent force determined the derivative gain through the leg kinematics. An actuated ratchet prevented restoring forces during stance. Physical granular media experiments were performed with glass beads in a bed prepared between jumps. |
Tuesday, March 16, 2021 1:06PM - 1:42PM Live |
F14.00009: Collective transport in an entangled robotic worm blob Invited Speaker: Saad Bhamla Through forming collectives, living and non-living systems can enable new functionalities that are not accessible to the individual. Most commonly studied swarms involve individuals that do not contact each other (fish school or drone swarms). Here, we develop a robophysical model of the collective dynamics of physically entangled collectives, worm blobs, where thousands of centimeter-long aquatic worms knot their bodies to form large-scale aggregates capable of surviving extended desiccation, mechanical impacts and directed locomotion under temperature gradients. The model consists of six 3-link, 2 motor planar robots (smarticles), with arms covered in barbs and mesh to aid entanglement. Using this robotic blob, we examine the relative importance of behaviors we observe in the living system: mechanical interactions (entanglements), differentiation of roles in the collective, and the existence of binders (force traction). The mean displacement of the robotic blob increased from 3.7±2.5 (all crawl) to 10.5±4.9 mm as the robots used the combination of three gaits. Our results reveal that gait differentiation is critical for collective movement, but synchronization is not required, and reduced activity of certain robots enhances the physical entanglement among individuals. |
Tuesday, March 16, 2021 1:42PM - 1:54PM Live |
F14.00010: Toward a Simple Model for Friction Dominated Multi-legged Locomotion Dan Zhao, Shai Revzen Legged locomotion is often modeled with non-slip contacts, in part due to the complexity associated with the number of possible Coulomb friction states that could arise. We have previously shown that multi-legged animals and robots almost inevitably slip a great deal, and that an alternative, non-Coulomb friction anzats provided promising simulation results with far less complexity. Here we report on advances in this model, where we have removed pitch and roll as dynamic parameters, thereby making the results dependent only on the ratios of leg stiffnesses and the ratios of friction coefficients, but not their absolute values. The new Python code takes 274+/-13 ms on a PC to simulate 500 time steps, running at x50 or more faster than real-time on one PC core. The error for predicting travelled distance per cycle (72 cycles, 14 trials, 7000 data points) is 7.3% (RMS). Per-cycle heading error was 0.13 rad (RMS). After finding and removing a systematic error in turning rates, per-cycle heading error reduced to 0.02 rad (RMS). We further illustrate the problems that arise solving for the Coulomb friction case. Overall, We hope this new simplified model will help us get a better understanding of multi-legged locomotion and multi-contact mechanics. |
Tuesday, March 16, 2021 1:54PM - 2:06PM Live |
F14.00011: Enhancing Legged Robot navigation of Rough Terrain via Use of a Tail Daniel Soto, Daniel I Goldman Robot locomotion across complex terrain has been widely studied for cases where the robot is much larger than obstacles or when the obstacles are much larger than the robot (e.g. climbing). Less is known about locomotion in environments where obstacles are of similar size to the locomotor; the most studied schemes rely on vision to map the terrain for appropriate footholds. Here we take a different approach, augmenting the emergent high-performance capabilities of a compliant-legged robot with a simple controllable tail. The quadruped robot (L=27cm, m=2.8kg, limb length 8cm) uses a diagonal couplet gait to locomote at 0.15 BL/sec on flat ground. When challenged by a 160 cm long 80cm wide terrain composed of blocks with heights chosen from an inverted Gaussian distribution centered around the limb length, the robot suffers failures consisting of limb jamming. Controlling the tail via periodic taps improves performance, allowing the robot to successfully traverse the terrain in 2/46 trials via controlled disturbances to decrease probability of jamming. Inspired by this, we implement a controller which senses high current draw associated with limb jams which further improves performance (3/25 trials successful). |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2025 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700