Our goal is to develop a motion planning framework that is applicable to large-scale robotic systems, and capable of handling complex task specifcations ranging from low-level kinematic/dynamic constraints to high-level, abstract requirements.
A robot and a workspace is modeled as a multi-body system that is composed of rigid bodies and joints. A motion planning problem is mathematically formulated as a constraint-satisfaction problem (CSP) in which constraints may have different levels of priority.
A special gradient-based algorithm is developed that can solve a large-scale prioritized CSP rapidly to meet tight real-time requirements of robotic applications. Up to now, the method has been applied to target-reaching task of a robotic arm, task scheduling of a humanoid robot, and real-time trajectory planning of a bipedal robot.