Proprioceptive-Based Multimode State Estimation for Wheeled-Quadruped Robot

This paper presents a novel multimode proprioceptive state estimation framework integrated with motor encoders and an Inertial Measurement Unit (IMU) for wheeled-quadruped robot operating over long distances. The standard assumption of using non-slipping stance legs for state estimation in legged robots becomes invalid for wheeled-quadruped platforms due to their hybrid wheeled-legged morphology. To address this issue, a probabilistic slip model that accounts for the additional displacement caused by wheeled motion is proposed. Additionally, by constructing an offset wheeled-legged contact model, we estimate the contact position and velocity to reduce the accumulated errors resulting from inaccurate contact point estimation. Based on these foundations, we integrate wheel rotation measurements with the displacement induced by the wheeled configuration and employ a Kalman filter for state estimation. By incorporating reliable contact constraint into the observation model, our approach simultaneously compensates for vertical drift while maintaining estimation accuracy. To accurately assess wheel-ground contact status, a generalized momentum-based normalized method for contact detection is applied. The cumulative errors caused by premature or delayed landings of the stance legs are thereby minimized. Finally, our multimode state estimation methods show superior performance in indoor and outdoor scenarios, achieving at least a 1.75 times reduction in the Root Mean Square Error (RMSE) across all directions compared to existing methods.