DROID-ANCHOR: Odometry-Anchored Recurrent Metric Depth Estimation
RoboticsComputer Vision and Pattern Recognition
Summary
The authors present Metric-DROID, a system designed to help robots estimate exact distances using just one camera and their own movement sensors. They combine visual information with odometry (movement data) to fix the usual problem where a single camera can't tell the true scale of the environment. Their approach uses a special neural network to process odometry data and account for its errors, making the distance estimation more reliable. They also introduce a method to fine-tune the system without losing what it already learned about geometry.
Authors
Yuxuan Chen, Brook Du
Abstract
Precise metric depth estimation is fundamental for autonomous robot navigation, yet monocular systems inherently suffer from scale ambiguity and scale drift. While recent recurrent flow-based SLAM systems have demonstrated state-of-the-art robustness, they remain scale-ambiguous. In this paper, we propose Metric-DROID, an end-to-end recurrent architecture that anchors visual SLAM to physical reality by integrating proprioceptive odometry. Our framework introduces the following innovations: (1) A LSTM Update Operator that encodes high-frequency odometry sequences into spatial feature maps, providing a persistent metric bias for iterative refinement. (2) An Uncertainty-Aware Metric Backend ($BA_{odom}$) that treats odometry as a geometric anchor with learned heteroscedastic covariance. By regressing a time-varying metric uncertainty $Σ_{o}$, our system intelligently balances visual re-projection and metric translation residuals, effectively mitigating the impact of wheel-slip and sensor noise. (3) We further propose a selective residual fine-tuning strategy to preserve pre-trained geometric priors while enabling zero-shot metric alignment.