VLD: Visual Learned Distance as an Interface for Reinforcement Learning Navigation
VLD: Visual Learned Distance as an Interface for Reinforcement Learning Navigation
This research introduces Visual Learned Distance (VLD), a scalable framework for image-goal navigation that decouples perception learning from policy learning. The authors train a self-supervised distance-to-goal predictor on approximately 3,000 hours of internet-scale and simulated video data. An RL policy is then trained in simulation using privileged geometric distances perturbed by GeoNoise, a learned noise model that mimics the errors of the distance predictor. At deployment, the policy uses VLD predictions zero-shot without retraining. The study reports that with VLD inputs, GeoNoise-trained policies achieve 58.7% success on HM3D and 76.2% on Gibson, compared to 4.9% and 11.5% for policies trained with generic noise. On a physical robot, the VLD-driven policy succeeds in 88% of trials, outperforming a strong end-to-end image-goal baseline that achieves 60% success.
The paper introduces Visual Learned Distance (VLD), a framework that decouples perception from policy learning to improve sim-to-real transfer in robot navigation, reporting specific success rates on HM3D, Gibson, and physical trials.
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.
Reported numbers
hours of internet-scale and simulated video
≈3,000 hours
View original evidence
al navigation that decouples perception learning from policy learning. We first train a self-supervised distance-to-goal predictor on about 3,000 hours of internet-scale and simulated video, providing a distance signal that a reinforcement learning (RL) policy can minimize. The RL policy is trained separately in simulation using privileged geometric distances perturbed by GeoNoise, a learned noise model that reproduc
Open source S3
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
al navigation that decouples perception learning from policy learning. We first train a self-supervised distance-to-goal predictor on about 3,000 hours of internet-scale and simulated video, providing a distance signal that a reinforcement learning (RL) policy can minimize. The RL policy is trained separately in simulation using privileged geometric distances perturbed by GeoNoise, a learned noise model that reproduc
Open source S3
es the viewpoint- and overlap-dependent errors of the trained distance predictor. At deployment, the policy consumes VLD predictions zero-shot, without retraining, inheriting ``where to go'' from large-scale visual training while retaining the robust navigation behaviors learned in simulation. This error modeling is key to the transfer: with VLD inputs, GeoNoise-trained policies reach $58.7\%$ success on HM3D and $76
Open source S4
.2\%$ on Gibson, whereas policies trained with generic temporally correlated noise collapse to $4.9\%$ and $11.5\%$. On a physical robot, the VLD-driven policy succeeds in $88\%$ of trials, compared with $60\%$ for a strong end-to-end image-goal baseline that outperforms it in simulation. Our decoupled design thus provides an alternative and scalable path toward reliable navigation policies that transfer from simulat
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org