Collision avoidance using contact information with multiple objects by multi-leg robot

In robotics, a walking through motion is complex because of the presence of multipoint contact objects in the working environment of a robot. To simplify the walking through motion of a robot, a virtual impedance field is implemented to the contact points of the robot and an object so that the robot avoids the object passively. The traveling direction of the robot is altered by a virtual repulsive force obtained from the position of the estimated obstacle and the virtual impedance field. The resulting action depends on the parameter of virtual impedance coefficients. Because a combination of parameters includes many things, reinforcement learning is employed to obtain an optimal motion. The optimization of the multipoint contact walking through motion of a robot is finally achieved by evaluating the walking motion while encountering complex obstacles in a dynamic simulator. The motion is implemented on a hexapod robot, and the results demonstrate the effectiveness of the proposed method.