Task allocation is an important problem in multi-robot system which can be defined with different setup for different application, i.e. coverage, surveillance and mining mission in static or dynamic scenarios. Our focus in this paper is exploring environment to accomplish tasks distributed over the environment by minimizing overall cost of the system. This problem is defined as a NP-Hard problem, thus will be more challenging in larger environments containing many robots and tasks. To solve multi-robot task allocation in very large environment we propose a new deployment-based framework. Our proposal divided the problem into two sub-problems: region partitioning and routing problem. This decomposition eases considering our problem specification in multi-robot system which are not easily considerable in other approaches, i.e distribution of the tasks or robots’ initial position. Load balancing is done globally by deploying robots in a proper location of the environment and assigning sub-regions among them. Sub-regions contains set of points, where the goal is visiting all the points individually by one of the robots. On the other hand, after deploying the robots, routing techniques can be simply applied to find shortest and safest paths for every robots. To search for solutions in this NP-hard problem, two methods are built on a tailor-made multi-objective scheme of Genetic Algorithm (GA) with a different setup and search operators, and a reinforcement learning approach. Simulation results testify the performance of our methods in comparison to existing ones.

任务分配是多机器人系统中的一个重要问题，可以为不同的应用定义不同的设置，即静态或动态场景中的覆盖、监视和挖掘任务。我们在本文中的重点是探索环境，通过最小化系统的总体成本来完成分布在环境中的任务。这个问题被定义为一个 NP-Hard 问题，因此在包含许多机器人和任务的更大环境中将更具挑战性。为了解决超大环境中的多机器人任务分配，我们提出了一种新的基于部署的框架。我们的提议将问题分为两个子问题：区域划分和路由问题。考虑到我们的多机器人系统中的问题说明，在其他方法中不容易考虑，即任务的分配或机器人的初始位置. 通过在环境的适当位置部署机器人并在其中分配子区域来全局完成负载平衡。子区域包含一组点，其中的目标是由一个机器人单独访问所有点。另一方面，在部署机器人后，可以简单地应用路由技术为每个机器人找到最短和最安全的路径。为了在这个 NP 难题中寻找解决方案，两种方法建立在具有不同设置和搜索运算符的遗传算法 (GA) 的定制多目标方案以及强化学习方法上。仿真结果证明了我们的方法与现有方法相比的性能。