基于渐进式神经网络的六足机器人避障策略迁移

Obstacle avoidance strategy migration of hexapod robot based on PNN

六足机器人直接在现实环境中进行避障训练,会出现数据采样效率低、样机与障碍物产生碰撞造成零件出现不可逆损伤等情况。采用迁移学习中的渐进式神经网络(PNN)来实现模型的多环境迁移。实验基于双重深度(Double-DQN)强化学习预训练模型,将预训练后的模型有机结合为PNN结构,进而完成从源任务到目标任务的避障策略迁移。根据仿真实验的结果显示,相较于其他设计方法,PNN学习目标任务花费的时间大大降低。然后将仿真器中训练好的PNN结构移植到六足机器人样机中测试,最终测试结果表明:六足机器人能够成功完成避障任务。

The hexapod robot directly conducts obstacle avoidance training in the real environment,and there will be situations such as low data sampling efficiency,and irreversible damage to parts caused by collisions between prototypes and obstacles.Progressive neural network(PNN)in migration learning is used to realize the multi-environment migration of model.The experiment organically combines the pre-training model based on Double deep Q-network(Double-DQN)into a PNN structure to complete the migration of obstacle avoidance strategies from the source task to the target task.According to the results of simulation experiments,compared with other design methods,the time taking for PNN to learn the target task is greatly reduced.Then the PNN structure trained in the simulator is transplanted to the hexapod robot prototype for testing.The final test result shows that the hexapod robot can successfully complete the obstacle avoidance task.