Using a Human-Like Double Support Force Distribution for Locomotion in Humanoid Robots

Locomotion is fundamentally the result of exchanging forces with the environment to generate movement in a desired direction. For bipeds, the most common locomotion is walking. It requires to control the forces applied in the environment to maintain balance and propel the body in the desired direction while respecting friction constraints. The relationship between the center of mass acceleration and contact forces is unique in the single support phase. On the contrary, in the double support phase, it is a non-deterministic problem. It is not uncommon to use heuristics to solve this problem such as minimizing a certain effort criterion. Given how humans regularly solve this problem it is interesting to take inspiration from human walking. The modified Twin Polynomial Method (mTPM) has shown results superior to the state of the art in understanding how humans distribute contact forces in double support [1]. In this paper, we show how human data differs from a ‘typical’ robot walk and propose using mTPM to generate a more human-like distribution for robots.