Pushing and pulling motion generation for humanoid robots using whole-body momentum control based on analytical inverse kinematics

As robotics technology advances, we expect humanoid robots to do strenuous work on our behalf. In everyday life, we frequently perform object manipulations. In particular, pushing and pulling motions are commonly done. Therefore, it is important to enable humanoid robots to perform these types of motions. Several studies on opening a door using a humanoid robot have been conducted. However, due to the delays of waiting for sensor data, the generated motions are slower than human motions. To speed up these motions, we propose two methods. One is to use prior knowledge of a predefined reaction force and to expand a preview control to handle the predefined force. The other is to accelerate the control of the total body momentum using analytical (not iterative) inverse kinematics. We verify the effectiveness of the proposed method in both simulation and the real humanoid robot HRP-4.

随着机器人技术的持续进步，我们期望人形机器人能够替代人类承担高强度作业。日常生活中，人类频繁开展物体操作任务，其中推拉动作是最为常见的操作类型之一。因此，使人形机器人具备此类动作的执行能力具有重要现实意义。目前已开展多项针对人形机器人开门作业的研究。但由于等待传感器数据存在时延，所生成的动作速度慢于人类自然动作。为加快此类动作的执行速度，本文提出两种优化方法：其一为利用预定义反作用力（predefined reaction force）的先验知识，并扩展预览控制（preview control）方法以适配该预定义反作用力；其二为采用解析（非迭代）逆运动学（analytical inverse kinematics）方法，加快全身动量的控制效率。本文通过仿真实验与真实人形机器人HRP-4，验证了所提方法的有效性。