MS-CRL: Multi-Scale Global Path Planning with Progressive Curriculum Reinforcement Learning
Nan Zhou, Xuqing Hu, Yixin Zhou, Rui Zhu, Fan Zhou, Ye Li, Guangqiang Yin
AI summary
Problem
Existing deep reinforcement learning methods for global path planning struggle with multi-scale map inputs due to inconsistent spatial representations across different map sizes and trajectory length variations that destabilize policy learning and hinder feature extraction.
Approach
The framework normalizes heterogeneous map inputs into a unified agent-centered representation, extracts cross-scale features via a global-local fusion network, and trains the policy using a progressive curriculum that gradually increases map size and obstacle complexity.
Key results
- Stabilized policy learning and training efficiency across varying map sizes
- Superior path success rate, quality, and planning efficiency over baselines
- Robust cross-scale adaptability via unified agent-centered observations
- First systematic solution for representation inconsistency and trajectory length discrepancies
Why it matters
It provides a scalable and stable foundation for autonomous navigation systems that must operate across diverse, dynamically changing environments without manual map reconfiguration.
Abstract
Global path planning provides high-level guidance for autonomous navigation, supplying reference paths for down- stream navigation and control modules. Deep Reinforcement Learning (DRL) has shown strong potential in this domain, but existing methods struggle with multi-scale map inputs. This limitation arises from inconsistent representations across different map sizes and trajectory length variations, which hinder feature extraction, destabilize policy learning. To ad- dress these challenges, we propose the Progressive Multi-Scale Curriculum Reinforcement Learning (MS-CRL) framework. MS-CRL incorporates a progressive curriculum reinforcement learning algorithm (ProgCRL) that mitigates instability from trajectory length discrepancies, a unified multi-scale represen- tation (UniMS) that normalizes spatial scales and resolves rep- resentation inconsistencies, and a Global-Local Fusion Network (GLFNet) that fully extracts both global and local features from the new representation for robust cross-scale policy learning. Extensive experiments on multi-scale map datasets demonstrate that MS-CRL enables effective global path planning, stabilizes policy learning, and achieves superior performance in path success rate, path quality, and planning efficiency, while signifi- cantly improving training efficiency and cross-scale adaptability compared with state-of-the-art baselines.