基于驾驶员模型的六足机器人自主/协同决策  

Hexapod robot self/collaboration decision based on the driver′s prior model

作　　者：陈潇磊 尤波[1,2] 李佳钰[1,2] 丁亮[3] 董正 Chen Xiaolei;You Bo;Li Jiayu;Ding Liang;Dong Zheng(Heilongjiang Provincial Key Laboratory of Complex Intelligent System and Integration,Harbin University of Science and Technology,Harbin 150080,China;Key Laboratory of Advanced Manufacturing Intelligent Technology,Ministry of Education,Harbin University of Science and Technology,Harbin 150080,China;State Key Laboratory of Robotics and System,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]哈尔滨理工大学黑龙江省复杂智能系统与集成重点实验室,哈尔滨150080 [2]哈尔滨理工大学先进制造智能化技术教育部重点实验室,哈尔滨150080 [3]哈尔滨工业大学机器人技术与系统国家重点实验室,哈尔滨150001

出　　处：《仪器仪表学报》2023年第4期91-100,共10页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金青年项目(51905136);国家自然科学基金面上项目(52175012);国家自然科学基金重点项目(91948202)资助。

摘　　要：重载六足机器人在野外地形环境移动作业时的决策智能水平亟待提高。然而,当机器人在尚未形成合理的决策结构层次时,直接采用其与环境进行交互方式进行常规的强化学习训练,将导致机器人的行为决策过于发散。因此,本文首先利用一种符合驾驶员决策逻辑的分步训练神经网络,得到驾驶员的决策经验模型,使机器人快速形成自主决策智能。此外,为融合人机决策优势,本文基于合作博弈理论,提出一种消除人机协同决策指令冲突的方法。搭建面向重载六足机器人人机协同决策的半物理仿真实验系统,开展实验的结果表明,机器人通过学习驾驶员先验模型和自主训练,其决策效果可接近驾驶员决策水平,同时人机协同决策指令可有效弥补单智能体决策指令的缺陷,在规则沟壑地形下协同决策指令的碰撞率指标优于驾驶员单智能体指令23.8%,障碍地形下协同决策指令的能量消耗指标优于机器自主单智能体指令34.1%。The level of decision-making intelligence of heavy-duty hexapod robots in the field terrain needs to be improved.However,if robots have not yet formed a reasonable decision structure level,the conventional decision-making reinforcement learning which is directly interact with the environment,will lead to the robot′s decision-making being too divergent.Therefore,this article first obtains the driver′s decision-making experience model through a step-training neural network which conforms to the driver′s decision-making habits.Hence,the robot can quickly form decision-making intelligence.In addition,to better play the advantages of human-robot decision-making,this article proposes a method to eliminate the conflict of human-robot coordinated decision-making commands based on the cooperative game theory.A semi-physical simulation experiment system for human-machine collaborative decision-making of heavyduty hexapod robots is designed and established.After carrying out experimental verification around the proposed methods,results show that the robot can approach the driver decision-making effect by learning the driver′s prior model and reinforcement training,and the effect of the human-robot collaborative decision-making commands can also make up for the defects in unilateral decision-making.In the regular ditches terrain,the collision index of the collaborative decision commands is 23.8%better than that of the single driver agent commands;in the obstacle terrain,the energy consumption index of the collaborative decision commands is better than that of the single robot agent commands by 34.1%.

关 键 词：六足机器人 协同决策 驾驶员先验模型 半物理仿真 神经网络 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置] TH39[自动化与计算机技术—控制科学与工程]