Reproduce “Dynamic Q-planning for Online UAV Path Planning in Unknown and Complex Environments” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lidia Gianne Souza da Rocha, Kenny Anderson Queiroz Caldas, Marco Henrique Terra, Fabio Ramos, Kelen Cristiane Teixeira Vivaldini
Researchers at Federal University of São Carlos, University of São Paulo, and University of Sydney developed a dynamic Q-planning framework for online UAV path planning that intelligently adjusts Q-Learning training iterations based on environmental complexity. The approach achieved 100% path completeness with lower computational cost and faster planning times compared to fixed-iteration Q-Learning and classical algorithms like A*, RRT, and PSO, in both indoor and outdoor unknown environments.
Key insights
- The empirical selection of Q-Learning iterations can be replaced by a dynamic, complexity-aware approach to improve real-time performance and reliability.
- Tracking the stability of Q-Learning rewards during training provides a robust indicator for optimal convergence, preventing both under and over-training.
- Integrating path smoothing techniques like cubic splines is essential to transform discrete reinforcement learning outputs into practical, executable trajectories for UAVs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.