SFA-MOAHA: A multi-strategy self-learning algorithm for multi-robot path planning in complex environments

Multi-robot systems are crucial in intelligent manufacturing. In dynamic environments, they face complex tasks requiring efficient coordination, with the multi-robot path planning (MRPP) problems being particularly significant. The current algorithms for MRPP problems face difficulties in dealing with dynamic uncertainties and real-time adaptability. To tackle the MRPP problem, this study proposes a spiral forward-adaptability multi-objective artificial hummingbird algorithm (SFA-MOAHA). It combines a spiral approximation strategy and a forward fusion learning strategy to boost exploration and exploitation, and integrates Deep Q-Network (DQN)to adaptively select the best search schemes based on iterative feedback. A multitude of simulation experiments have been carried out to validate the efficacy of the SFA-MOAHA. The experiments cover tests with three benchmark test sets (ZDT, DTLZ, and WFG), engineering design problems, and solving MRPP problems with different robot numbers and map complexities. Compared to traditional algorithms, SFA-MOAHA shows superiority in obtaining well-distributed and converged solutions for the benchmark tests and engineering problems. In two environments of different complexities, SFA-MOAHA shortens the path lengths of 4 mobile robots by 6.23% and 7.45% respectively, and those of 8 robots by 7.89% and 9.61% respectively, outperforming multi-objective artificial hummingbird algorithm (MOAHA). This study demonstrates that SFA-MOAHA works excellently in complex situations and offers an adaptive learning solution for real-time multi-robot coordination.