Towards High-Fidelity On-Board Attitude Estimation for Legged Locomotion via a Hybrid Range and Inertial Approach

Abstract

Legged robots display a characteristically periodic motion. Measuring and tracking this motion has traditionally been performed using general inertial measurement techniques. While widely applied in robotics, this approach is limited in dynamic legged locomotion due to the excessive accumulation of drift from severe impact shocks (nearly 9 g in single leg experiments). This paper introduces the attitude estimation problem for legged locomotion and shows preliminary results from a more powerful combined range and inertial sensing approach. Based on a modified Extended Kalman Filter the method uses ground-directed range sensors, the stride period, and other periodic features of legged locomotion in order to address inertial drift. Together this provides rapid, robust estimates of flight phases and attitude necessary for extended dynamic legged operations.