Foot Contact Estimation for Legged Robots in Rough Terrain

Abstract

Accurate sensing of the terrain shape is a key issue for robots moving in rough environments. For legged robots, an important aspect is the terrain inclination for each foot which is in contact with the ground. Knowledge about the terrain inclination is required to design controllers that mitigate the risk of slippage by minimizing the tangential forces applied on the ground. In this paper, we introduce a method to estimate the contact surface normal for each foot of a legged robot relying solely on measurements from the joint torques and from a force sensor located at the foot. The force sensors at the foot optically determine the deformation of the foot to estimate the force applied to the sensor (optical force sensors). We formulate a measurement model of the optical force sensor and combine it with the joint torque measurements in an Extended Kalman Filter (EKF). The resulting method is able to determine the contact force and surface normal through contact of the foot with the ground. The approach is implemented and evaluated on a torque-controllable quadrupedal robot and shown to reliably estimate the surface normal during dynamic motions for each leg individually.