Balance improvement of a humanoid robot while walking

Walking is a daily movement employed since the dawn of time by most humans but falls may nevertheless occur. Research linked to the management and the avoidance of falling are still carried out since we do not yet understand how the human brain recognizes falling prediction patterns, and reacts accordingly. These knowledge of human balance would strengthen the robustness of biped humanoid robots and especially bio-inspired robots. Since a robot that falls can harm not only its environment but also itself, implementing a balance system becomes crucially relevant. This will allow the robot to interact in various environments where perturbations could be encountered. However, humanoid robots are far from reaching human balance abilities since they are frequently unstable during locomotion due to their relatively high centre of mass and limited support polygon. This master thesis is devoted to the development of a balance controller as complement to the existing bio-inspired controller for the compliant robot COMAN, in simulation [Van der Noot, 2017b]. Two main goals guide this work. Firstly, perturbations must be detected to trigger the use of the additional balance controller. And secondly, the implementation of a balance controller to face disturbances during walking is achieved. The detection of perturbation is realized by comparing specific data of the robot with a reference robot that is not facing perturbations. If the error between these data exceeds a certain threshold, then a perturbation is sensed. During the tests, all perturbations are sensed after around 0.12s. Concerning the implemented balance controller, it relies on a force-based controller and allows to improve the performance of the robot against lateral pushes. Explanations on the implemented methods, detailed results, comparisons with human responses and discussions are provided in this report.