State Estimation for Point-Foot Parallel-Legged Bipedal Robot

Bipedal robots have shown great potential in academic, medical and military fields. Bipedal robot motion control requires accurate acquisition of body state as feedback information, so it is important to accurately obtain body state information in real time. For parallel-legged point-foot robots, complex structural characteristics bring many challenges to state estimation. On the one hand, the continuous collision and impact of the point-foot with the ground causes the inertial measurement unit (IMU) to generate high-frequency noise. On the other hand, the increase in the number of joints in parallel structures brings more kinematic parameter noise and joint encoder noise. This paper proposes a real-time state estimation framework for XingT, a parallel-legged bipedal robot. Specifically, the five-link forward kinematics information is used as the Extended Kalman observation state to update the IMU prior state. In addition, for foot contact state, this paper applies a contact probability model and integrates foot height, gait phase and other information to obtain the contact probability. The proposed method was successfully applied to the XingT robot. The results show that the method is more robust than the linear Kalman Filter and can operate at a higher frequency (400 ~Hz).