论文部分内容阅读
为了研究仿人、能量高效的双足机器人步行,研制了由MACCEPA(mechanically adjustable compliance and controllable equilibrium position actuator)柔性驱动器驱动的半被动双足机器人,并实现了其动力学仿真系统。提出一种基于再励学习的步行控制方法。该方法首先采用Q-学习方法学习机器人在理想环境中的稳定步行步态及其控制策略,然后将此步态和控制策略作为模糊优胜学习方法的参考步态和参考控制策略并在线学习模糊网络的优胜值参数。仿真结果表明:利用学习训练的结果控制柔性驱动器在步行相转换时的动作,机器人可以实现稳定动态步行。
In order to study the walk of biped robot with energy efficiency and human-like ability, a semi-passive biped robot driven by MACCEPA (Flexible Adjustable and Controllable Equilibrium Position actuator) was developed and its dynamic simulation system was realized. A walking control method based on re-energizing learning is proposed. The method first learns the stable walking gait and its control strategy in the ideal environment by Q-learning method, and then uses this gait and control strategy as reference gait and reference control strategy for fuzzy winning learning method and online learning fuzzy network The winning parameter. The simulation results show that the robot can achieve steady dynamic walking by using the result of learning training to control the motion of the flexible driver during walking phase conversion.