From Depth to Safe Depth: Uncertainty-Aware Monocular Perception for Mobile Robot Navigation
From Depth to Safe Depth: Uncertainty-Aware Monocular Perception for Mobile Robot Navigation
This research paper introduces a lightweight plug-in module that transforms predictions from standard monocular depth foundation models into 'safe depth' estimates by incorporating uncertainty. The authors demonstrate that while off-the-shelf models like Depth Anything V2 lead to high collision rates (63%) in simulated navigation tasks due to distance over-estimation, their proposed method reduces collisions to 3% and improves navigation success from 37% to 70%. The approach uses a frozen depth model with a small correction head trained via conformal calibration to adjust depth values pixel-wise, effectively handling challenging conditions like glass surfaces and texture-less walls.
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.
- Environment
- SimulationOpen source S2
Reported numbers
episode count
120 episodes
View original evidence
arXiv:2611.13769v1 Announce Type: new Abstract: Monocular depth foundation models make it tempting to replace depth sensors on mobile robots with a single RGB camera. We show that, used off the shelf, this is unsafe: in a simulated indoor navigation benchmark, a standard Nav2 stack driven by Depth Anything V2 (metric) collides in 63\% of 120 episodes, because the model systematically over-estimates the distance to ne
Open source S2depth error AbsRel
0.262
View original evidence
is set by conformal calibration on a held-out world, and uncertainty also decides which pixels may be treated as floor. The correction halves the depth error (AbsRel $0.262\rightarrow0.134$) and reduces near-range over-estimation from 59.8\% to 12.0\%. In navigation, safe depth raises the success rate from 37\% to 70\% and lowers collisions from 63\% to 3\%, close to the 2\% of ground-truth depth. A constant safety m
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2611.13769v1 Announce Type: new Abstract: Monocular depth foundation models make it tempting to replace depth sensors on mobile robots with a single RGB camera. We show that, used off the shelf, this is unsafe: in a simulated indoor navigation benchmark, a standard Nav2 stack driven by Depth Anything V2 (metric) collides in 63\% of 120 episodes, because the model systematically over-estimates the distance to ne
Open source S2
is set by conformal calibration on a held-out world, and uncertainty also decides which pixels may be treated as floor. The correction halves the depth error (AbsRel $0.262\rightarrow0.134$) and reduces near-range over-estimation from 59.8\% to 12.0\%. In navigation, safe depth raises the success rate from 37\% to 70\% and lowers collisions from 63\% to 3\%, close to the 2\% of ground-truth depth. A constant safety m
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org