Balancing control

Inverse dynamics approach

Balancing the robot is a very difficult task in the domain of humanoid and biped robots regarding the very small region of support. It is less challenging in quadrupped robots as they can easiy form new support polygons by their feet to capture center of Mass (CoM) and maintain the balance. Keeping balance becomes more challenging when we expect the robot to be compliant. We have developed an inverse dynamics based controller which takes care of full dynamics of the robot and generates joint torques which satisfy our targets. With this controller which is formulated as a per-timestep Quadratic optimization, we can track cartesian trajectories for whatever point on the robot. This multi-contact formulation is generic, real time and flexible, using slack variables.


Long demonstrations on the real robot

URL is outside of EPFL