A robophysical study of active force sensing for least-resistance traversal of cluttered large obstacles

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.