MamAM: A Novel Deep Reinforcement Learning Architecture for Multi-Robot Task Allocation

This study presents a novel approach for efficiently solving multi-robot task allocation(MRTA) problems, which include tasks characterized by deadlines and workload, as well as constraints related to the robots' working capacity.We propose a method called the Mamba Attention Model (MamAM), which consists of two main components: an encoder and a decoder. The encoder, based on the Mamba Network, is responsible for capturing the global dependencies among task points. The decoder, utilizing an attention network, decodes the encoded information provided by the encoder to generate the task selection sequence for each robot. We use a policy-based reinforcement learning approach to train the MamAM model. Finally, we demonstrate the effectiveness of the MamAM model through a series of comparative experiments. The findings reveal that, compared to non-learning-based methods for solving this problem, the MamAM model performs better in terms of inference speed and task completion rate.