Abstract:
The Central Pattern Generator for Humanoid Robots based on Preview Control has certain limitations at which the system becomes unstable or unsolvable. Two of these corner cases are lack of actuation (reduction in the number of actuated degrees of freedom) and Zero-Moment Point moving outside the ground contact polygon. A Decentralized Pattern Generator is presented in this study to overcome these limitations and the suggested approach is developed as a multi-rigid-body dynamics solution which is a step towards finding a necessary condition for humanoid robot stability instead of a sufficient condition, as suggested by the humanoid robotics research community.
The formulation of the pattern generation problem as a multi-body dynamics problem is a key part in deriving the DPG for under-actuated humanoid robots. A number of robot models with fully-actuated and under-actuated configuration are analyzed and motion generation is performed on 5 different cases in this study. Rigid-body dynamics and multi-body dynamics are used to incorporate the robot hardware along with the choice of joint actuators in the motion generation process for the robot. An optimal control system topology for the DPG is also presented in this study.
The DPG uses an iterative approach for solving the equations of motion and joint forces/torques under the position and velocity constraints imposed on the robot hardware. An under-actuated humanoid robot NUSTBOT-3 is designed and developed with un-actuated ankles and a ROS based solution incorporating the decentralized pattern generator is implemented to overcome the limitations of the central pattern generator. The multi-body dynamics simulation and analyses are performed to verify the solutions generated by the decentralized pattern generator.
