Belief Space Planning for Robust Humanoid Locomotion

We have implemented a BSP (Belief Space Planning) controller inspired by human walking behaviour's neurophysiological mechanisms on the PAL Robotics REEM-C humanoid robot. BSP controllers assuming maximum likelihood have been previously successfully utilized by our group for various similar demanding applications. The posture and locomotion performance of the REEM-C robot with BSP controllers has been experimentally tested and validated in the Eurobench@IIT humanoid testing facility by means of the benchmarking framework and software already developed in Eurobench. The main purpose of the work reported here is to assess if and how the application of a BSP controller is viable and useful in the humanoid robotics domain. The BSP walking controller has shown remarkably good performance in tracking tasks. However, it has not performed well on slopes in simulation. A new more advanced approach aiming to cope with slopes and unstable terrain is under development.