Reproduce “Learning Coverage Paths in Unknown Environments with Deep Reinforcement Learning” on OpenResearch

About this paper

Arvi Jonnarth, Jie Zhao, Michael Felsberg

Researchers at Linköping University and Dalian University of Technology developed an end-to-end deep reinforcement learning framework for online coverage path planning in unknown environments. This system uses continuous control signals, multi-scale maps with a novel frontier encoding, and an incremental total variation reward to achieve efficient and robust coverage, outperforming several classical and RL-based baselines.

Key insights

• End-to-end deep reinforcement learning with continuous control effectively enables adaptive and flexible online coverage path planning in unknown environments.
• The combination of multi-scale maps and frontier map encoding is critical for representing large environments efficiently while retaining necessary long-term planning information.
• An incremental total variation reward is crucial for encouraging agents to fill small, remaining uncovered regions, significantly improving the completeness of coverage.

Summary by alphaXiv.