VLD: Visually Learned Distance as an Interface for Reinforcement Learning Navigation
What happened
Lazar Milikic, Manthan Patel, and Jonas Frey propose the Visually Learned Distance (VLD) framework, which separates perception learning from policy learning for image-goal navigation: first, a distance-to-goal predictor is trained self-supervised on approximately 3,000 hours of internet and simulation video; then, a reinforcement learning policy is trained in simulation using GeoNoise to perturb privileged geometric distances; at deployment, VLD predictions are used zero-shot without retraining. The authors report that on a physical robot, the VLD-driven policy achieved an 88% trial success rate, compared with 60% for an end-to-end image-goal baseline that performed better in simulation.
AI 根据报道生成 · 2 days ago更新
Reporting timeline
Follow the reporting to explore different perspectives.
- arXiv cs.ROVLD: Visual Learned Distance as an Interface for Reinforcement Learning Navigation
The authors propose the Visual Learned Distance (VLD) framework, which decouples perception learning from policy learning for image-goal navigation: first, a distance-to-goal predictor is trained self-supervised on about 3,000 hours of internet and simulated video; then, in simulation, a reinforcement learning policy is trained using privileged geometric distances perturbed by GeoNoise; at deployment, VLD predictions are used zero-shot.
Activity for this event
- Current
- 3
- Peak
- 10Nov 9 14:00
- Past 24 hours
- -50%
The trend compares participants observed throughout the same period, so its range may be smaller than the current interest count. Move or tap to inspect each hour; use the arrow keys on a keyboard.