Reproduce “Anatomical Landmark-Guided Deep Reinforcement Learning for Autonomous Gastric Navigation” 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
Haoxuan Wu, Sishen Yuan, Haitao Gao, Zhen Li, Xiuli Zuo, Hongliang Ren
A framework for autonomous gastric navigation using wireless capsule endoscopy (WCE) employs anatomical landmark-guided deep reinforcement learning, achieving 97.3% mean gastric coverage in simulation and 87% in ex-vivo pig stomachs. This approach reduced procedure time by 53% compared to manual control in ex-vivo settings and improved simulation coverage by 53.8% over unguided DRL agents.
Key insights
- Basing DRL policies on universally conserved anatomical landmarks, rather than raw visual data, significantly enhances transferability across diverse patient anatomies and reduces computational load.
- The combination of a simplified, low-dimensional state representation and an optimized reward function enables DRL agents to efficiently achieve high gastric mucosal coverage.
- Robust sim-to-real transfer is achievable by decoupling trajectory planning from real-time execution and employing an adaptive controller to counteract physical discrepancies and dynamic environmental factors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.