Design Attention Awareness Among Robots for Uncertainty-Adaptive Heterogeneous Teaming
Chao Huang, Rui Liu · 2022
Attracted by team size and robot function diversity, heterogeneous multi-robot system (MRS), where multiple robots with different capabilities are coordinated to perform tasks, has been widely used for large-scale scenarios, including disaster search and rescue, site surveillance, and social security. However, the availability of robot members is influenced by real-world factors such as being far away from the target position, occupied by other missions, and robot failure. In a decentralized scenario without human supervisory, due to the availability of robot members is uncertain, it is challenging to flexibly team robots with complementary capabilities to optimally balance reasonable task allocations and maximum utilization of robot capability. To solve this problem, in this paper, a novel flexible teaming method, inner attention (innerATT), is developed to flexibly team available robot members to execute tasks as task types and available robot members change. innerATT is designed by integrating a novel cooperation likelihood calculation method into a multi-agent actor-critic reinforcement learning architecture. With the innerATT, robot availability and capability will be analyzed to calculate the cooperation likelihood among robot members. Based on the cooperation likelihood, robots can flexibly form teams to meet task requirements. Scenarios with different task varieties ("Single Task", "Double Task", and "Mixed Task") and different available robot members were designed. The effectiveness of the innerATT was validated by its accuracy in flexible cooperation.