Imitation of Human Motion on a Humanoid Robot Using Inverse Kinematics and Path Optimization

Abstract

Robots are mainly used in dull, dirty, and dangerous environments where it is difficult for the humans to work. Robot imitations present a natural approach for humanoid robots to mimic the action of humans in real time. However, it is not an easy task to build an efficient and robust imitation system because of the high degree of freedom involved in motions. Even direct assignment of observed human motion to robot's joints is not possible due to the physical differences between human and humanoid robots. In this paper we present an off-line inverse kinematics based human mimicking system to map human upper limbs motion to robot's joints in a smooth and safe manner. For recording the human motion, any sensor capable of capturing the joint angle trajectories can be used. In our work, we capture the human data with Kinect V1 and converted depth image of human into skeleton form to get joint positions. In addition, optimization of data is done using both single and multi objective optimization techniques. Experimental results show that the system is robust and efficient enough to imitate various human motions.