APPA: An adaptation path planning algorithm for real-time obstacle avoidance in multi-robot systems

摘要

With the rapid advancement of robotics technology, the deployment of multi-robot systems in logistics, industrial automation, and public services has expanded significantly. However, complex and dynamic environments present stringent requirements for the real-time performance, stability, and coordination capabilities of path planning algorithms. To address these challenges, this paper introduces an Adaptive Path Planning Algorithm (APPA) that enables exploration-based real-time obstacle avoidance in multi-robot systems. The APPA employs a hierarchical two-level planning architecture: in the first stage, an incremental search algorithm is used to generate a potential path framework; in the second stage, an adaptive exploration mechanism refines the path dynamically to adapt to environmental changes and moving obstacles. To enhance both planning efficiency and coordination, APPA integrates a reward mechanism that balances goal orientation and obstacle avoidance. We also introduce UCBG, an exploration strategy that combines KDE with multi-armed bandit theory for intelligent action selection. This approach uses KDE to model reward distributions without assuming specific parametric forms. Unlike traditional UCB methods that rely on Gaussian assumptions, our method captures uncertainty through entropy-based exploration, improving the exploration-exploitation balance in multi-robot systems. Experimental results show that APPA significantly outperforms baseline algorithms such as ORCA, ALAN, and PRIMAL. While deep reinforcement learning approaches like ALAN excel in static environments through neural network-based policy learning, APPA’s non-parametric uncertainty modeling provides superior adaptability in dynamic scenarios with unpredictable obstacle patterns, achieving 94% success rate compared to ALAN’s 52% in high-density dynamic environments. APPA attains a 35% reduction in average pathfinding time compared to PRIMAL.