We propose an approach to find low-makespan solutions to multi-robot multi-task planning problems in environments where robots block each other from completing tasks simultaneously. We introduce a formulation of the problem that allows for an approach based on greedy descent with random restarts for generation of the task assignment and task sequence. We then use a multi-agent path planner to evaluate the makespan of a given assignment and sequence. The planner decomposes the problem into multiple simple subproblems that only contain a single robots and a single task, and can thus be solved quickly to produce a solution for a fixed task sequence. The solutions to the subproblems are then combined to form a valid solution to the original problem. We showcase the approach on robotic stippling and robotic bin picking with up to 4 robot arms. The makespan of the solutions found by our algorithm are up to 30% lower compared to a greedy approach.

翻译：我们提出一种方法，用于在多机器人在环境中相互阻塞无法同时完成任务的场景下，寻找多机器人多任务规划问题的低完工期解。我们引入该问题的形式化描述，使得基于带随机重启的贪心下降法生成任务分配与任务序列成为可能。随后使用多智能体路径规划器评估给定分配与序列的完工期。该规划器将问题分解为多个仅含单机器人与单一任务的简单子问题，从而可快速求解以产生固定任务序列的解。子问题的解被组合后形成原始问题的可行解。我们在最多包含4个机械臂的机器人点画与机器人抓取任务中展示了该方法。与贪心方法相比，我们算法找到的解的完工期最多降低30%。