Perception-less Terrain Adaptation through Whole Body Control and Hierarchical Optimization

This paper presents a control approach based on a whole body control framework combined with hierarchical optimization. Locomotion is formulated as multiple tasks (e.g. maintaining balance or tracking a desired motion of one of the limbs) which are solved in a prioritized way using QP solvers. It is shown how complex locomotion behaviors can purely emerge from robot-specific inequality tasks (i.e. torque or reaching limits) together with the optimization of balance and system manipulability. Without any specific motion planning, this prioritized task optimization leads to a natural adaption of the robot to the terrain while walking and hence enables blind locomotion over rough grounds. The presented framework is implemented and successfully tested on ANYmal, a torque controllable quadrupedal robot. It enables the machine to walk while accounting for slippage and torque limitation constraints, and even step down from an unperceived 14 cm obstacle. Thereby, ANYmal exploits the maximum reach of the limbs and automatically adapts the body posture and height.