Chair: Shai Revzen

Abstract:
Multi-legged robots with six or more legs have designs with great stability and maneuverability using a low number of actuators. With more legs in contact with the ground, it becomes harder and unnecessary to ensure that they all move in compatible ways to ensure non-slip contact, and slipping becomes inevitable in our robots. Modeling multi-legged motion with slipping and producing reliable predictions of body velocity is challenging and computationally expensive. In this work, we investigated an algorithm that allows us to efficiently model body velocity and ground contact forces given robot shape and shape-changing velocity. This algorithm relies on previous experimental observations showing that even while slipping, multi-legged robots are principally kinematic, where body velocity could be computed through a linear local connection from the shape-changing velocity. This tool scaled well with an increasing number of legs without losing accuracy. We applied our algorithm towards modeling multiple robots with different morphology and different gaits, together with predicting 3-dimensional ground contact forces. Further extension of this model enabled its usage on stairs or slopes. The simplicity of the model led to easy parallelization and motion planning of the robot’s behaviors on GPU. Analysis of the modeling residual term led to a simple and data-efficient sim-to-real transfer to unmodeled robot dynamics.