Search papers, labs, and topics across Lattice.
This paper introduces Metric-DROID, an innovative recurrent architecture that integrates proprioceptive odometry with visual SLAM to achieve precise metric depth estimation for autonomous navigation. By employing a novel LSTM Update Operator and an Uncertainty-Aware Metric Backend, the system effectively anchors visual data to real-world metrics, addressing the challenges of scale ambiguity and drift. Key results demonstrate significant improvements in depth accuracy and robustness against sensor noise and wheel-slip, showcasing the potential for enhanced navigation in complex environments.
Anchoring visual SLAM to real-world metrics could revolutionize depth estimation, eliminating scale ambiguity and enhancing navigation reliability.
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 $危_{o}$, 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.
BAIR