Research On Mission-driven Intelligent Decision-making Method Of Robot Operation

Flexible intelligent manufacturing is a hotspot in the field of manufacturing and robots replace human beings to realize automation and intelligence of manufacturing process.However,current production planning,such as task decomposition,robot operation and robot scheduling,often requires manual prior decision-making.Aiming at the complex and changeable production tasks,this paper studies the task-driven intelligent decision-making method for robot operation.According to the task and the current working environment information,the decision-making of robot operation and its action sequence is obtained.The main contents are as follows:The product information and the status information of workpiece shape,size,position and posture in production system are described hierarchically.A data structure and its XML representation method of robot task based on workpiece status are designed and established.The database of robot,workpiece status and working environment is designed and established to provide necessary information support for subsequent decision-making.Based on HTN programming theory,an automatic generating planning domain knowledge algorithm for tasks is proposed.Combining the generated planning domain knowledge with the general planning domain knowledge,robot task planning is realized,and robot task is effectively decomposed into a set of executable operations.The reinforcement learning method is used to solve the scheduling problem of complex production system.Based on the production system model and optimization objective,a description method of state space,action space and action reward is proposed,and an intelligent decision algorithm of production system scheduling based on Q-Learning is established.In order to solve the problem of low efficiency of current algorithms,the Q-Learning algorithm based on case transfer is studied by introducing the idea of knowledge transfer.A method of calculating the similarity between case task and target task is proposed,and the mapping method of transferring case knowledge to current task decision-making is studied.Compared with Q-Learning algorithm,the learning performance of the proposed algorithm has been improved to varying degrees.Taking the actual production case as an example,task description method for robots,the established work information database,the robot task planning algorithm and the intelligent scheduling algorithm for production systems presented in this paper are used to described the production tasks and realize the robot task planning and intelligent scheduling for production systems based on the production work information,and the robot task simulation for the robot workstation is carried out.The experimental results verify the feasibility and effectiveness of the proposed method.