Multi-layered real-time controllers for humanoid's manipulation and locomotion tasks with emergency stop

This paper describes a practical method to construct real-time controllers to achieve locomotion and manipulation tasks with a humanoid robot. We propose a method to insert emergency stop functionality to each layer to avoid robot's falling down and joint overloads even if recognition and planning error exist. We explain implementation of multi-layered real-time controllers on HRP2 robot and application to several manipulation and locomotion tasks. Finally, we evaluate emergency stop functionality in several manipulation tasks.