Title:Sensing environmental interaction physics to traverse cluttered obstacles

Abstract:When legged robots physically interact with obstacles in applications such as search and rescue through rubble and planetary exploration across Martain rocks, even the most advanced ones struggle because they lack a fundamental framework to model the robot-obstacle physical interaction paralleling artificial potential fields for obstacle avoidance. To remedy this, recent studies established a novel framework - potential energy landscape modeling - that explains and predicts the destabilizing transitions across locomotor modes from the physical interaction between robots and obstacles, and governs a wide range of complex locomotion. However, this framework was confined to the laboratory because we lack methods to obtain the potential energy landscape in unknown environments. Here, we explore the feasibility of introducing this framework to such environments. We showed that a robot can reconstruct the potential energy landscape for unknown obstacles by measuring the obstacle contact forces and resulting torques. To elaborate, we developed a minimalistic robot capable of sensing contact forces and torques when propelled against a pair of grass-like obstacles. Despite the forces and torques not being fully conservative, they well-matched the potential energy landscape gradients, and the reconstructed landscape well-matched ground truth. In addition, we found that using normal forces and torques and head oscillation inspired by cockroach observations further improved the estimation of conservative ones. Our study will finally inspire free-running robots to achieve low-effort, "zero-shot" traversing clustered, large obstacles in real-world applications by sampling contact forces and torques and reconstructing the landscape around its neighboring states in real time.