Floating-base state estimation is a critical research area for humanoid robots since their free movement in space makes it difficult to determine their precise location and orientation. Accurately estimating the floating-base state enables an MPC controller to generate optimized motion trajectories that are safe and feasible for the robot to execute. This is especially important in situations where the robot is subjected to external disturbances or operating on uneven terrain.
However, there are several challenges to floating-base state estimation, including noisy sensory data from IMUs and blurred vision data due to task-purpose impacts. Additionally, the nonlinear and tightly coupled dynamics of the robot’s base and limbs make accurate estimation a significant challenge. Addressing these challenges is critical to improving the accuracy and robustness of floating-base state estimation and enabling effective control strategies for robotic systems.