Multi-mechanical-arm multi-target searching training method and training device

The invention belongs to the technical field of mechanical arm design, and provides a multi-mechanical-arm multi-target searching training method and device. The training method comprises the steps that all mechanical arms are matched with all targets based on a clustering algorithm, and track planning is conducted on all the mechanical arms based on a fixed track planning algorithm so that all the targets matched with all the mechanical arms can be found; the planned track is used as pre-training experience of each corresponding mechanical arm; the pre-training experience serves as priori knowledge accumulated by the corresponding mechanical arm, and interactive iteration training between the mechanical arm and a target is conducted on the basis of a reinforcement learning algorithm till a preset iteration threshold value is reached; further obtaining each trained mechanical arm; wherein the reward function in the reinforcement learning algorithm comprises a first reward function, a second reward function an