Hyperbolic Maps and Objective Preserving Rewards for Learning Coverage Paths in Unknown Environments
MCML Authors
Matthias Schubert
Prof. Dr.
Core PI
Abstract
Matthias Schubert
Prof. Dr.
Core PI
Abstract
Coverage path planning is the task of nding paths that cover an entire area. This process is essential for robotic applications, such as lawn mowing or vacuum cleaning. While recent deep reinforcement learning approaches show promise, they often suer from structural ineciencies, leading to inecient paths when approaching high coverage. We address these limitations by introducing HOP-CPP, a novel framework with three key contributions: (1) a novel observation representation comprising a Hyperbolic Map that unies spatial and historical information into a single continuously distorted image, (2) an auxiliary reward that is objective-preserving under ideal conditions and eliminates fragmented coverage patterns near obstacles, and (3) wall-gliding as a low-level control mechanism that works together with the auxiliary reward to enable ecient obstacle-adjacent coverage. In benchmarking experiments, HOP-CPP outperforms current methods in coverage, episode length, and total turns.
inproceedings BS26
ECML-PKDD 2026
European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases. Naples, Italy, Sep 07-11, 2026. To be published.
Authors
J. Blake • M. SchubertResearch Area
BibTeXKey: BS26