Environmental exploration is one of the common tasks in the robotic domain which is also known as foraging. In comparison with the typical foraging tasks, our work focuses on the Multi-Robot Task Allocation (MRTA) problem in the exploration and destruction domain, where a team of robots is required to cooperatively search for targets hidden in the environment and attempt to destroy them. As usual, robots have the prior knowledge about the suspicious locations they need to explore but they don’t know the distribution of interested targets. So the destruction task is dynamically generated along with the execution of exploration task. Each robot has different strike ability and each target has uncertain anti-strike ability, which means either the robot or target is likely to be damaged in the destruction task according to that whose ability is higher. The above setting significantly increases the complexity of exploration and destruction problem. The auction-based approach, vacancy chain approach and a deep Q-learning approach based on strategy-level selection are employed in this paper to deal with this problem. A new simulation system based on Robot Operating System and Gazebo is specially built for this MRTA problem. Subsequently, extensive simulation results are provided to show that all proposed approaches are able to solve the MRTA problem in exploration and destruction domain. In addition, experimental results are further analyzed to show that each method has its own advantages and disadvantages.

环境探索是机器人领域中的常见任务之一，也称为觅食。与典型的觅食任务相比，我们的工作集中在探索和破坏领域的多机器人任务分配（MRTA）问题上，在该问题中，需要一组机器人协作搜索隐藏在环境中的目标并试图摧毁它们。像往常一样，机器人对需要探索的可疑位置具有先验知识，但他们不知道感兴趣目标的分布。因此，销毁任务是随着探索任务的执行动态生成的。每个机器人具有不同的打击能力，每个目标具有不确定的抗打击能力，这意味着在破坏任务中，机器人或目标都可能因其能力较高而受到损坏。上述设置大大增加了勘探和破坏问题的复杂性。为了解决这个问题，本文采用基于拍卖的方法，空缺链方法和基于策略级选择的深度Q学习方法。针对此MRTA问题专门构建了基于机器人操作系统和凉亭的新仿真系统。随后，提供了广泛的仿真结果，表明所有提出的方法都能够解决勘探和破坏领域中的MRTA问题。另外，对实验结果进行了进一步分析，表明每种方法各有优缺点。为此，本文采用了空缺链方法和基于策略选择的深度Q学习方法。针对此MRTA问题专门构建了基于机器人操作系统和凉亭的新仿真系统。随后，提供了广泛的仿真结果，表明所有提出的方法都能够解决勘探和破坏领域的MRTA问题。另外，对实验结果进行了进一步分析，表明每种方法各有优缺点。为此，本文采用了空缺链方法和基于策略选择的深度Q学习方法。针对此MRTA问题专门构建了基于机器人操作系统和凉亭的新仿真系统。随后，提供了广泛的仿真结果，表明所有提出的方法都能够解决勘探和破坏领域中的MRTA问题。另外，对实验结果进行了进一步分析，表明每种方法各有优缺点。大量的仿真结果表明，所有提出的方法都能够解决勘探和破坏领域的MRTA问题。另外，对实验结果进行了进一步分析，表明每种方法各有优缺点。大量的仿真结果表明，所有提出的方法都能够解决勘探和破坏领域的MRTA问题。另外，对实验结果进行了进一步分析，表明每种方法各有优缺点。