Dynamic traversal of large gaps by insects and legged robots reveals a template

Sean W. Gart, Changxin Yan, Ratan Othayoth, Zhiyi Ren, Chen LiarXiv 2019 · arXiv

Abstract

Navigating complex 3-D terrain remains a significant challenge for mobile robots. While traditional planning relies on high-fidelity sensing, animals often utilize high-speed dynamics to overcome obstacles. This study investigates how discoid cockroaches and bio-inspired hexapedal robots traverse large gaps. By developing a simplified mathematical "template" model, researchers identified that momentum and body pitch are primary determinants of success. Applying these insights to a robot with an active tail resulted in a 50% increase in gap-crossing capability, demonstrating that dynamic traversal can bypass the need for complex neural feedback in challenging environments.

Introduction

In mobile robotics, navigating unstructured environments such as building rubble or forest floors is a formidable task. Traditional "deliberate" planning methods utilize sensors like LiDAR or vision to map paths. However, in real-world scenarios, these sensors often prove too slow or noisy for rapid movement.

Examples of complex 3-D terrain including forest debris (a), cracked wood (b), dry earth (c), and rubble (d).

Animals frequently employ "dynamic traversal," using kinetic energy to overcome obstacles without pausing for complex computation. This research seeks to uncover the physical principles allowing legged systems to cross large gaps—a common 3-D obstacle—using passive mechanics and high-speed dynamics.

Methods

The researchers employed a comparative approach involving animal trials, robotic experiments, and mathematical modeling:

  1. Animal Trials: Discoid cockroaches (Blaberus discoidalis) were filmed crossing gaps ranging from 0.2 to 1.0 body lengths at speeds up to 12 body lengths per second.
  2. Robot Trials: A six-legged, RHex-class robot served as a physical model, operated "open-loop" (without sensors) to isolate the role of pure mechanics.
  3. The Template Model: A reduced-order dynamic model was developed, treating the body as a rigid ellipsoid rotating about a posterior hinge. Lagrangian mechanics were used to predict if the "head" would reach the far side before falling.
  4. Active Control: An active motorized tail was added to the robot to test the model's predictions regarding body pitch control.

The study generated original experimental datasets including 420 cockroach trials, 110 robot trials, and 90 trials with the tail-equipped robot.

Key Insights and Results

The study revealed that traversal success is primarily predicted by the "Head-Bridge" factor: whether the animal's head reaches the far edge before the body pitches down excessively. While crossing gaps larger than half a body length is nearly impossible at slow speeds, high approach speeds provide the necessary "flight time."

Key findings include:

  • Performance Parity: Both cockroaches and the open-loop robot successfully traversed gaps up to 1.0 body length.
  • Pitch Dynamics: Initial body pitch and pitch angular velocity are critical, as they "buy time" for the body to move forward before gravity pulls the head below the ledge.
  • Engineering Application: Using the template's insights, the researchers programmed the robot's tail to flick backward, transferring angular momentum to the body. This increased the robot's maximum traversable gap from 0.8 to 1.2 body lengths—a 50% improvement.

Discussion and Future Work

While the "template" model accurately predicts the bridging phase, it does not yet account for the "gripping" phase. Cockroaches excel at recovery due to their flexible bodies and spiny legs, whereas the rigid robot often fell backward after reaching the far side.

Future research will focus on "anchor" models that incorporate leg-terrain interactions and body flexibility. This "terradynamics" approach is being expanded to other obstacles, such as large bumps, to help robots not only reach the other side but successfully "stick the landing" in complex terrains.

Summary updated

BibTeX
@misc{gart2019dynamic,
  author = {Sean W. Gart and Changxin Yan and Ratan Othayoth and Zhiyi Ren and Chen Li},
  title = {{Dynamic traversal of large gaps by insects and legged robots reveals a template}},
  year = {2019},
  eprint = {1911.02526},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1911.02526},
}
Figure