Model-Based Imitation Learning For Monocular Vision-Based Navigation
Model-Based Imitation Learning For Monocular Vision-Based Navigation
This research presents a hybrid approach combining model-based planning with imitation learning for reliable navigation using a single monocular camera. The method employs a learned collision model to predict worst-case obstacle clearance distributions, which inform trajectory optimization via collision-risk costs. Human demonstrations are used to balance collision risk with task objectives and to optimize algorithmic parameters like warm starts. Experiments demonstrate highly reliable navigation in cluttered environments across different robot embodiments, achieving a substantial reduction in collision rates compared to recent end-to-end and model-based baselines.
The paper proposes a model-based imitation learning framework that combines learned collision models with human demonstrations to achieve reliable monocular vision-based navigation, reporting reduced collision rates compared to end-to-end and model-based baselines in cluttered environments.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Source：arXiv Robotics — research abstracts · arxiv.org