The aim of this project is to develop an unified framework of autonomous and continuous locomotion and manipulation in a constrained environment by robots which can work in repetitive, tedious, or a heavy task for humans in large-scale manufacturing.
Due to a huge search space and many constraints in such an environment, it is hard to find a feasible motion.
In order to solve this problem, the motion libraries of locomotion and manipulation will be prepared from the environment primitives and the whole body trajectories corresponding to enriched contact conditions (e.g. sliding) from a virtual environment in a dynamics simulator in advance by using a sort of machine learning.
Then the continuous locomotion and manipulation without interruption will be archived by executing environment measurement/recognition, a whole-body motion planning and its controller in frequent cycle from the motion libraries and sensing data.
A whole-body controller will be also developed in order to adapt a real-environment by considering measurement and unmodeled errors under the constraints by introducing the knowledge of non-linear control schemes.
In this way, we propose a new framework of locomotion and manipulation for robots to expand the reachable area and acquire the skills to perform new tasks.