基于脉冲强化学习和CPG的四足机器人分层运动控制

The control technology of quadruped robots has gradually developed, and quadruped robots have been widely applied in fields such as rescue, military, and exploration. In the application of quadruped robots, how to effectively reduce energy consumption without sacrificing control performance is an important bottleneck problem. In response to this problem, this paper proposes a hierarchical control algorithm based on the spiking reinforcement learning(SRL) algorithm and the central pattern generator(CPG). First, considering that spiking neurons have lower energy consumption than traditional artificial neurons, this paper constructs a SRL algorithm based on a spiking neural network(SNN) and uses it as the control center. Then, when the action space of the control task is too large, it is difficult for the SRL algorithm to achieve good control effects, thus the CPG model is used as the low-level controller, and the SRL algorithm is utilized to accept state information and adjust CPG parameters for controlling the movement of the quadruped robot. Finally, the SRL-CPG control algorithm is experimentally verified on the quadruped robot model Gbot built in the Webots environment. The results show that the SRL-CPG control algorithm can be effectively applied to the motion control of quadruped robots and significantly reduce energy consumption.