Physics-based Human Walking Animation





Physics-based human animation is realistic because the motion follows the laws of physics. Given sufficient computational power and robust control systems, simulation also allows on-line interactivity. But the difficulty of programming control systems makes the motion sometimes look mechanical.  On the other hand, animation generated from motion capture data preserves most of the details of human motion, but is not always easily modified when on-line interaction is required. Both physics-based simulation and motion capture have advantages and drawbacks, and we hope to combine these two techniques to generate human animation that is realistic, appealing and interactive.


We choose walking motion as our testbed for this approach because it is a common human activity that has many stylistic variations.  To combine motion capture and simulation, we first model the human body as an articulated system composed of rigid links and rotary joints. The model is driven by joint torques, and the corresponding movements of its links are generated by forward dynamics simulation. We then constructed a control system for walking, which generates appropriate torques based on the robot's dynamic states and environment condition. The control system was designed based on our physical intuitions, robot control theories, and biomechanics studies.  Given a walking control system, we can then incorporate motion capture data into our system to enrich the styles of motion.  We have built a simulation system that can generate steady planar walking motion of the lower body.  We are now working on three-dimensional walking motion and improving the naturalness of the motion.


Links to results

         Planar walking motion by tracking MOCAP data ( the figure with only lower body is the simulated motion )

         Planar walking motion without MOCAP data

         3D walking motion without MOCAP data