On The Reinforcement Learning Based Task Allocation Of Multi-robot

In recent years, the field of application of the robot expands constantly, Monomer robot can no longer meet the people's needs in many areas. Research and use multi-robot through synergies and cooperation to complete certain tasks, becoming a new kind of robot application form, and it is also attention by academic circles at home and abroad increasingly.Multi-robot coordination is a very important part during the research of multi-robot systems, which multi-robot task allocation (Multi-robot Task Allocation: MRTA) occupy a major position. Robot groups decompounds reunification goal into individual tasks performed by single robot, and doing a reasonable allocation, which is multi-robot assignment. The main application areas of MRTA is for a number of occasions, such as service robots, dangerous environment detectors, education and entertainment systems, automatic construction and so on. It can be foreseen that the application of MRTA will be of great change in the community, It also can greatly improve the quality of people'lives as well as modernization of industrial , agricultural national defense.This paper researches the arithmetic of MRTA system . First it studies a market-based method to realize the MRTA in a complex dynamic environment; and then researches on a fire-fighting mission to study task allocation arithmetic based on reinforcement learning, to achieve the hidden collaboration between robots, a reinforcement learning arithmetic that the award value attenuation with the time is then proposed. at the same time, Vicsek model is applied to self-organizing control strategy of swarm robots, and achieving the cluster of swarm robots acts. Finally, designing a fully functioning multi-robot simulation system ,it can be used to simulate and test different types of tasks, it provides a common test platform for the theory study on large-scale multi-robot systems. The details are as follows:(1) Research on the distributed cooperation method based on the market mechanism in distributed task-allocation of multi-robot system. That is how to allocate tasks to robots in multi-robot system without centralized control and meet optimization and reliability at the same time to enable the robot to a higher efficiency accomplish their tasks.Firstly, the multi-robot fighting mathematical model is established, at the same time ,considered that the multi-robot system is fully distributed, a distributed cooperative task-allocation method for fire-fighting missions in multi-robot based on the market mechanism is designed, so as to shorten task completion time, raise the completion of the task ,mainly responsible for the completion of the distribution of tasks, tasks in the implementation of the robot, such as the conversion of the state to solve the robot in a time-constrained environment of dynamic task allocation mandate. The main innovation of this chapter are the establishment of multi-robot fire-fighting task's mathematical model, introducing the distribution of tasks in the market mechanisms, so as to further optimize the distribution of results; designing a dynamic bidding formula including the distance fitness, the fire emergency, and also the time of urgency solve a robot in the implementation of the distribution of tasks which is easy to neglect the issue of time constraints.(2) The traditional robot collaborative learning algorithm for a multi-task robots are more based on reinforcement learning, but this method will be a large state space and a slow convergence when there are a large number of robots. It's proposed to use the reinforcement learning arithmetic that incentives value attenuation with the passage of time, which indirectly completes Implicit coordination mechanism between robots, different robot will be after the fire to the different sports in different convergence strategy at the first time, to maximize the completion of the task as much as possible. Each robot will do the separate training, therefore to reduce the system's state space, improving the convergence rate of study. The results of simulation show that the convergence and effectiveness of the arithmetic, achieve robots autonomous coordination in the fire-fighting mission The arithmetic improves the robot tasks allocation strategies, raises the learning speed compared with multi-robot learning methods, and reduces the state space.(3) In the practical application, a number of low cost robot can be carried through joint communications detection on the map to complete the task of mine clearance. So if swarm robots could through simple partial interaction formation of clusters can be greatly reduce the volume of traffic under the premise of achieved the purpose of the assignment system, while increasing robustness. In this paper, the Vicsek model is applied to acts of self-organizing control strategy of swarm robots, designing direction convergence module, collision avoidance module and aggregation module, using robots to achieve communication between regional groups of local clusters of robots. Simulation results show the effectiveness of control strategies.(4) Developed distributed multi-robot simulation systems based on the C / S structure, designed its groups architecture and individual architecture respectively, achieve the distribution and collaboration of multi-robot systems, simulate various functions of the actual multi-robot systems. Using the simulation system to simulate a variety of environments in this paper, and run them using the following methods: achieve the environment display and overall status monitoring at the server-side, at the client-side analog simulation with the other entities. The simulation process use the Agent-oriented approach to the systems analysis, and use the object-oriented technology to design and implement. Take into account that JAVA is Platform-independent and Algorithm unrelated therefore, to adopt JAVA of the Development and Implementation. According to actual operation, the simulation system is good open, interactive, real-time and with reliability, and provide a common test platform to organize multi-robot system reasonably.Summarily, this paper accomplishes some foundational theory research work for the Multi-robot Task Allocation arithmetic, including dynamic task allocation method and task allocation arithmetic based on reinforcement learning. However there are some other problems to be researched and resolved, such as improved task allocation arithmetic dependents on the conditions of communication, the proof of convergence for reinforcement learning arithmetic.