Abstract: The loose cooperative group robot system has broad application prospects in the current coal mine auxiliary robot operation. However, the task assignment process of the loose cooperative group robot system did not provide feedback to the division process, resulting in insufficient efficiency and rationality of the task division and assignment process. To address this issue, an efficient task assignment algorithm for coal mine underground group robots based on an improved Rubinstein negotiation strategy is proposed. Based on the multi-party game features of task division and assignment in group robot systems, the Rubinstein negotiation strategy is extended from a bipartite game to a multi-party joint game. A "bid-bargain-counteroffer" rule for multi-party negotiation games is proposed. From the perspective of the difference between the execution capability and task execution status of individual robots, a discount factor calculation method based on the task completion quantity per unit time of robot individuals is proposed. A task completion status feedback parameter model based on the task execution status of each assignment cycle is also proposed to achieve dynamic task division and assignment. By collaborating with three groups of robots to perform overall monitoring tasks in coal mining areas, experimental verification is conducted on the performance of the algorithm. The results show the following points. ① Algorithm 3 uses an improved Rubinstein negotiation strategy. Algorithm 1 directly uses the ratio of the number of unmanned aerial vehicles in each group multiplied by their running speed as the standard for task division and assignment in three groups of unmanned aerial vehicles. Algorithm 2 uses the Rubinstein negotiation strategy of multi-party negotiation without considering the feedback parameters of task completion status. Algorithm 3 has a higher efficiency in task division and assignment than Algorithm 1 and Algorithm 2 by 30.10% and 18.29% respectively. ② The average maximum time difference for the three groups of unmanned aerial vehicles based on Algorithm 3 to execute tasks is 42 seconds. It is 77.66% and 65.29% optimized compared to Algorithm 1 and Algorithm 2, respectively. This is because Algorithm 3 introduces task completion status feedback parameters to timely evaluate the task execution process of the task participants. Algorithm 3 provides feedback on the task assignment and execution process to the task division stages, making the task division and assignment more accurate.