Bulletin of the American Physical Society
APS March Meeting 2024
Monday–Friday, March 4–8, 2024; Minneapolis & Virtual
Session G38: Robophysics IIIFocus
|
Hide Abstracts |
Sponsoring Units: DBIO DSOFT Chair: Daniel Goldman, Georgia Tech Room: 103D |
Tuesday, March 5, 2024 11:30AM - 12:06PM |
G38.00001: Affordances of Animals and Machines Invited Speaker: Daniel E Koditschek Over the past two decades, robophysicists’ understanding of the physical interactions between bodies and the work they can perform within specified environments, has greatly facilitated the design of environmentally informed robot architectures as well as helped biologists hypothesize environmentally shaped animal designs. This talk is intended as a call to arms for robophysicists to lend their great insights and scientific prowess toward the converse problem of affordances – architecturally informed identification of task-exploitable environmental features. I will present a brief account of work in my lab over the past decade that attempts to develop an abstracted, compositional view of environmental features suitable for reasoning about their possible role as affordances to be exploited by appropriate compositions of templates - abstracted behavioral features of a robot’s architecture. This presently geometric view of the environment would be greatly enhanced by annotations of substrate mechanics whose development has been pioneered by robophysicists. The talk will conclude with speculative remarks on how such a collaborative merger might offer new approaches to the study of animal innovation while at the same time help generate more innovative robot behavior in extra-terrestrial environments that no animal has yet encountered. |
Tuesday, March 5, 2024 12:06PM - 12:18PM |
G38.00002: Geometry of contact: contact planning for multi-legged robots via spin models Baxi Chong, Di Luo, Tianyu Wang, Gabriel Margolis, Zhaocheng Xu, Massimiliano Iaschi, Pulkit Agrawal, Marin Soljacic, Daniel I Goldman While gaits for bipedal and quadrupedal robots have been extensively studied, multi-legged locomotors with six or more legs have received comparatively less attention and are often limited to symmetric alternating tripod gaits. The complexity of the gait search space increases exponentially with each added leg, challenging our conventional intuitions developed from bipedal and quadrupedal templates. This study develops an analytic tool to explore the high-dimensional search space in multi-legged systems. Combining geometric mechanics theory, graph theory, spin model, and robophysical experiment, we study the locomotion of the general multi-legged locomotors with a focus on the contact planning and its coordination with internal shape changes. Using geometric mechanics analysis, we map the multi-legged contact planning design into a well-defined graph problem. Leveraging the symmetries in locomotion, we map the graph problem to special cases of spin models. This mapping facilitates the identification of global optima in polynomial time, enabling a systematic analysis of multi-legged robots grounded in physics principles. We verified our models using robophysical experiments on a 20-cm long hexapod robot: our analysis identified an agile asymmetric gait with speed of 0.62 ± 0.03 BL/cyc (body length per cycle), faster than a standard symmetric alternating tripod gait with 0.43 ± 0.02 BL/cyc. |
Tuesday, March 5, 2024 12:18PM - 12:30PM |
G38.00003: Characterizing Mechanical Properties of Natural Deformable Substrates with a Direct-Drive Robotic Leg John Bush, Yifeng Zhang, Jake Futterman, John Ruck, shipeng liu, Ethan Fulcher, Douglas Jerolmack, Kenton Fisher, Ryan C Ewing, Feifei Qian Deformable substrates cover the majority of Earth and planetary surfaces. The mechanical strength and rheological properties of these substrates play a pivotal role in determining traversability, hillslope stability, and planetary exploration success. However, it is often difficult to accurately and rapidly assess substrate strength and rheology in situ. In this study, we created a direct-drive (i.e., gearless) robotic leg with the ability to accurately sense contact forces and measure terrain characteristics, without the need for additional sensor payload. We show from lab and field measurements that the robotic leg could discern subtle variations in mechanics quantities such as penetration resistance, yield stress, brittleness, and resilience. The ability to detect these variations enabled characterizing and understanding the distinct rheological behaviors exhibited by different types of natural terrain materials. Our study highlights the potential for legged robots to use their locomotive limbs as novel terrain sensors, and represents a key step towards the development of terrain-aware legged robots that can map terrain properties by walking. |
Tuesday, March 5, 2024 12:30PM - 12:42PM |
G38.00004: Gait switching enables body pitch modulation during legged burrowing in granular media Amber Young, Laura K Treers, Hannah S Stuart Subsurface exploration is accomplished by few legged robotic systems, yet performed by numerous animals. In our previous work, a mole-crab inspired robot (EMBUR 1.0) self-burrowed vertically by excavating granular media with counter-rotating leg pairs. EMBUR 1.0 had two modes of burrowing: one where its body pitch increased from its initial value, and a second mode where its pitch decreased from initial. These modes limit the achievable burrow depth. We note that the real mole crab maintains a more intermediate body pitch.We hypothesize that EMBUR can regulate its body pitch during burrowing to increase the achievable burrow depth and reliability. This work introduces EMBUR 2.0—an updated self-burrowing robot capable of body pitch control. A simulation based in 2D granular resistive force theory predicts unstable regions for operation, and suggests methods for modulating pitch by changing the direction of motion of each leg pair. Gait switching allows EMBUR 2.0 to adjust its pitch to recover from an increasing or decreasing body pitch mode. We characterize how interchanging between three gaits at varying body pitch thresholds influences burrowing trajectory. This work shows that discontinuous gait changes improve performance, and should be considered in legged self-burrowers. |
Tuesday, March 5, 2024 12:42PM - 12:54PM |
G38.00005: Tactile feedback enhances multi-legged locomotion on rugged landscapes Juntao He, Baxi Chong, Zhaochen Xu, Esteban Flores, Daniel Soto, Daniel I Goldman Organisms leverage various sensory modes when locomoting through complex environments, from long-range visual cues to short-range tactile feedback. In cluttered and crowded environments, long-range visual sensing becomes challenging. Instead, rapid localized response (short-range feedback) is presumed more effective. In recent robotic applications, short-range tactile sensing on feet improved locomotion in complex terrain via decentralized intra-leg coordination. However, it remains unclear if foot-level tactile information could contribute to overall locomotion via high-level centralized, intra-leg processing. To address this, we developed a centipede-like multi-legged robophysical model (L= 60 to 160 cm, 3 to 8 segments each with four degrees of freedom for limb and body movement) with point-like feet and tested it on rugose terrains. We found that adjusting the amplitude of a vertical body wave assisted in countering local heterogeneities and improved open-loop locomotion on the different terrains. We then developed a “binary” tactile sensor to detect ground contact at each foot and integrated this onto the model to estimate local terrain complexity to inform a centralized controller that determines suitable vertical waves. This improved speed by 50% over rough terrains as well as reduced variance by 60%, indicating that the robot was less impacted by the environmental “noise”. This approach to tactile information for multi-legged locomotion can improve our understanding for how such sensory cues affect biological systems when long range sensing is unavailable. |
Tuesday, March 5, 2024 12:54PM - 1:06PM |
G38.00006: Combined machine and human learning facilitates fast robot turns on steep granular slopes. Malone L Hemsley, Deniz Kerimoglu, Daniel Soto, Joseph S Brunner, Sehoon Ha, Tingnan Zhang, Daniel I Goldman
|
Tuesday, March 5, 2024 1:06PM - 1:18PM |
G38.00007: Proprioceptive sensing to aid with locomotion adaptation in mud shipeng liu, Shravan Pradeep, Sen Gao, Douglas Jerolmack, John Bush, Siyuan Meng, Jiaze Tang, John Ruck, Feifei Qian Moving on natural muddy terrains such as river beds, forest floors, and nearshore is extremely challenging, as subtle changes in mud composition and water content can lead to significant differences in mechanical behaviors. To enable terrestrial robots to effectively determine mud properties and flexibly adapt their locomotion strategies accordingly, in this study we explore proprioceptive terrain-sensing methods for robots to estimate substrate mechanical properties through joint torques during locomotion. We performed highly-controlled shear and penetration experiments with systematically-varied mud properties and found that mud resistance forces depended sensitively on both water content and clay-to-sand ratio. Interestingly, when scaled by a "distance from jamming'' variable, all shear force measurements from the high clay content regime collapsed to a non-dimensional master curve. This suggested that the distance from jamming could be used to determine the mechanical properties for a wide range of natural cohesive terrain materials. We demonstrated that based on the distance from jamming estimated from its limb joints, a flipper-based terrestrial robot could adapt its locomotion strategies and robustly move through muddy substrates with widely-varied properties. |
Tuesday, March 5, 2024 1:18PM - 1:30PM |
G38.00008: Anisotropic body compliance facilitates robotic sidewinding in complex environments Velin H Kojouharov, Tianyu Wang, Matthew Fernandez, Jiyeon Maeng, Daniel I Goldman Sidewinding, a locomotion strategy characterized by the coordination of travelling waves of lateral and vertical body undulation (referred to as the gait “template”), is observed in certain desert snakes and has been successfully reconstructed by limbless robotic systems for effective movement across diverse terrestrial terrains. Inspired by control simplification achieved via mechanical intelligence in lateral undulation [Wang et al 2023], which offloads feedback control to passive body mechanics, we developed a new kind of sidewinding limbless robophysical model. This 105 cm long device features a bilateral cable actuation system that resembles organism muscle actuation mechanisms and allows for programmable anisotropic body compliance. By varying the body compliance while sidewinding, we observed that with appropriate directional compliance the robot achieved both a lower cost of transport on hard ground and improved navigation through heterogeneities in lab (peg board) and outdoor terrains relative to experiments without compliance. When sidewinding through heterogenous environments, compliance helped minimize the effect of impeding robot-environment interactions, allowing the robot to 1) squeeze through obstacles or 2) brush by them without large changes in body orientation. |
Tuesday, March 5, 2024 1:30PM - 1:42PM |
G38.00009: A robophysical study of active force sensing for least-resistance traversal of cluttered large obstacles Yaqing Wang, Ling Xu, Chen Li To traverse cluttered large obstacles, animals and robots must transition across various locomotor modes. Our recent work on cockroaches and their robophysical models showed that such transitions are strenuous barrier-crossing transitions across basins of a potential energy landscape. Because a potential energy landscape's gradients are conservative forces, we hypothesize that a robot can sense its obstacle contact forces to infer landscape gradients, reconstruct the landscape, and seek saddle points between basins to transition with least resistance. As the cockroach oscillates its head up and down during transition, we further hypothesize that head oscillation is a form of active sensing and facilitates the robot's force sensing. Here, we tested these hypotheses in our model system of a robot traversing grass-like beam obstacles, by measuring obstacle contact forces while oscillating its head at various frequencies. By assuming Coulomb friction, we obtained normal obstacle contact forces and found that they matched landscape gradients well. The reconstructed landscape matched the ground truth, and the match increased with head oscillation frequency. These findings supported our hypotheses. Our next step is to develop a saddle-seeking algorithm for least-resistance transitions. |
Tuesday, March 5, 2024 1:42PM - 1:54PM |
G38.00010: Making every step an experiment: proprioceptive sensing during locomotion for enhanced mobility and data collection in earth and planetary explorations Yifeng Zhang, Ethan Fulcher, Diego J Caporale, shipeng liu, John Bush, John Ruck, Daniel E Koditschek, Douglas Jerolmack, Feifei Qian For legged robots moving through complex, deformable terrains, substrate reaction forces "felt" through locomotive appendages can be more informative than visual or other exteroceptive inputs for inferring environment properties and adapting locomotion strategies. Recent advancements in direct-drive actuators have enabled the development of individual robotic legs with the ability to measure contact forces and terrain characteristics through actuator phase current. The goal of our study is to extend this novel capability from stationarily mounted individual legs to dynamic quadrupedal robots, and develop platforms that can rapidly assess substrate strength and rheology during continuous locomotion. Here we report our results on two aspects of this development: (i) Implementing momentum-based observers to improve the accuracy of force sensing during dynamic motion, and (ii) understanding how different robot locomotive behaviors (e.g., leg touchdown speed and direction, body pose) could inform distinct terrain characteristics. Going forward, these understandings could enable legged robots to select custom locomotive behaviors to effectively gather environment information during every step, and use the information in turn to enhance their mobility during earth and planetary explorations. |
Tuesday, March 5, 2024 1:54PM - 2:06PM |
G38.00011: Gait design and mechanical intelligence facilitate open-loop limbless obstacle aided locomotion Tianyu Wang, Baxi Chong, Anushka Bhumkar, Velin H Kojouharov, Christopher J Pierce, Daniel I Goldman Limbless organisms, such as snakes and nematodes can exploit interactions with obstacles to enhance their mobility. This is referred to as obstacle-aided locomotion (OAL) and we posit can be robophysically modeled as a coordination of high-level self-deformation patterns (referred to as gait templates) alongside low-level adaptations of the body to diverse environmental conditions. In robotics, previous research on OAL has predominantly employed stereotyped traveling-wave gait templates (circular trajectory in the shape space) and active body deformations based on sensory feedback for obstacle navigation. In this work, we test the efficacy of elliptical gaits in OAL, using geometric mechanics to generate novel undulatory templates. By converting the body-obstacle contacts in the position space into constraints on lateral and rotational movements in the robot shape space, we discover that elliptical gait templates can generate greater forward thrust through interactions with a single post. When we implement the elliptical gait templates on a bilaterally actuated cable-driven limbless robot, we verify in robophysical experiments that, 1) the anisotropic compliance can enable the robot to move through lattices without the need of actively changing body control; 2) elliptical gaits with higher eccentricity excel in environments with lower obstacle density while lower eccentricity elliptical gaits are preferable in environments with higher obstacle density. |
Tuesday, March 5, 2024 2:06PM - 2:18PM |
G38.00012: Comparative biological and robophysical study of amphibious fishes moving on mud of variable strength Divya Ramesh, Gargi Sadalgekar, Hongbo Zhang, Jiangqi Tan, Dami Kim, Alex Nath, Qiyuan Fu, Zachary Souders, Lei An, Chen Li Amphibious fishes often encounter mud as they move across the water-land interface. Mud is challenging to move on, because it can either stay solid or flow like a fluid and the yield strength at which it goes through solid-fluid transition varies with how wet it is. Here we prepared mud of controlled, variable wetness states/strengths and systematically studied the mudskipper moving on them, which uses both pectoral fins in phase to “crutch” the body forward on hard ground. We also collected preliminary data for the bichir, which uses both pectoral fins alternately while laterally undulating the body on hard ground, and the ropefish, which relies on body undulation on hard ground. As mud weakens, all species sank more deeply and had a longer body section contacting mud during locomotion. The mudskipper’s performance (forward displacement per cycle) reduced, and it bent its tail to propel to compensate. The bichir’s and the ropefish’s performance did not change, and they lifted parts of the body, likely to reduce drag. To understand these observations, we are developing a robophysical model. Its body can laterally bend to propel against mud and vertically lift to control contact. Its two fins can move in or out of phase or be removed, to study a diversity of body-fin coordination. |
Tuesday, March 5, 2024 2:18PM - 2:30PM |
G38.00013: Mechanical intelligence facilitates limbless locomotion in cluttered aquatic environments Nishanth Mankame, Tianyu Wang, Matthew Fernandez, Christopher J Pierce, Daniel I Goldman While open-water limbless locomotion has been well-studied robotically and biologically, less research has been devoted to principles of limbless swimming in cluttered aquatic environments. Motivated by the remarkable control scheme simplification that mechanical intelligence (MI, purely passive, mechanically controlled body-environment interactions) offers limbless systems in highly damped environments [Wang et al. 2023], we hypothesize that MI could also play a role in cluttered aquatic locomotion. Our robophysical limbless model (L = 45cm), whose actuation is inspired by organism muscle actuation morphology, generates shape changes via programable directional compliance, generated by bilateral cable actuation and characterized by a parameter G. G=0 is a tightly controlled shape and increasing G leads to larger potential deviations from commanded shapes. During locomotion in a shallow water tank with uniformly spaced obstacles (9cm diameter posts, hexagonal pattern, 25cm spacing), for intermediate G, the robot’s performance was insensitive to wave amplitudes and spatial frequencies. Lower G control resulted in the robot bouncing off obstacles or jamming. Unlike in terrestrial locomotion in which kinematic efficiency was independent of temporal frequency, in this inertial regime, the robot moved most effectively (0.51 BL/cycle) at lower undulation frequencies (below 0.1 Hz). Thus, MI can be utilized for effective open-loop locomotion in cluttered inertial regimes. |
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. |
© 2024 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