AI-Assisted Motion Planning for Autonomous Robots Using Reinforcement Learning

The novel Hierarchical Reinforcement Learning (HRL) approach is introduced in the present investigation to aid in efficient and scalable motion planning in pursuit-evasion environments using autonomous robot swarms. The suggested system is designed with a two-tiered architecture: one layer for decentralized Multi-Agent Deep Deterministic Policy Gradient (MADDPG) for continuous navigation planning and another layer for centralized Double Deep Q-Network (Double DQN) for dynamic target allocation. The system uses a probabilistic ensemble-based interaction mechanism to evaluate epistemic and aleatoric uncertainty, as well as dynamically changes the job reassignment frequency depending on environmental unpredictability, to provide adaptive coordination. According to the experimental findings, the HRL technique is considerably more effective and flexible compared to traditional Deep Reinforcement Learning (DRL) methods. Particularly, HRL has a 91%-win rate by 20 agents as opposed to DRL’s 72%, so as the number of agents rises, it keeps making better decisions with less computing work. Regarding real-time multi-agent robots functioning in complex and variable settings, these results emphasize HRL’s resilience, flexibility, and feasibility.

Hali tarjima qilinmagan