BP networks based trajectory planning and inverse kinematics of a reconfigurable Mars rover

The inverse kinematics of series manipulators presents an inherent complexity due to their various structures and kinematics constraints. To a novel reconfigurable Mars rover's arm, the inverse kinematics had been solved by numerical method combined with space geometry relations, but the complex calculating process can not be utilized in real-time control. In this paper, some actions in common use are programmed in the child rover, and the BP neural network was used to solve the inverse kinematics and trajectory planning. To a desired trajectory, some solutions by direct kinematics algorithms and geometry relations corresponding to the trajectory were used as the BP network's training data set, and the non-linear mapping from joint-variable space to operation-variable space was obtained with iterative training and learning method. The results show the output of training is consistent with the desired path, and the obtained reference trajectory is smooth.