Year: 2026
Type: preprint
Abstract: Multi-robot collaborative task allocation and path planning are characterized by solution spaces that expand rapidly with problem scale and by increasingly conflicting objectives. Existing swarm intelligence algorithms often suffer from slow convergence and reduced diversity in later stages. To overcome these limitations, this paper proposes a hybrid algorithm, termed HIGWO-NSGAII, which integrates an improved grey wolf optimizer (GWO) with NSGAII. The algorithm first adopts NSGAII as the framework for non-dominated sorting and elitism preservation. A GWO-based gene injection mechanism is then designed for permutation encoding to accelerate convergence. In addition, Q-learning is introduced to adaptively schedule the crossover probability, mutation probability, and GWO guidance ratio, thereby dynamically balancing exploration and exploitation. Finally, a constrained 2-opt local search and an elite archive are employed to improve the quality of the Pareto front. Comparative experiments were conducted against seven algorithms in three grid-based scenarios. The results show that HIGWO-NSGAII achieved the best values in HV, IGD, Spacing, and the number of Pareto solutions. In both the small- and large-scale experiments, the IGD reached 0.0000, and the standard deviations of the key metrics were zero. In the medium-scale experiment, HV improved by 22.3% and IGD decreased by 90.3% relative to the second-best algorithm, with a Wilcoxon test p-value below 0.001. The proposed algorithm exhibits good convergence, stability, and scalability, providing an effective solution for multi-robot collaborative task planning.
Source: Research Square
Authors: Xiancheng Fan, Yongxia Wang, Xinyu Ling, zhuo Li
Institution: Anhui Institute of Information Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225652580