VLD: Visual Language Goal Distance for Reinforcement Learning Navigation

TL;DR

This paper introduces VLD, a scalable framework for goal-conditioned robotic navigation that leverages a self-supervised distance predictor trained on internet video data, enabling effective sim-to-real transfer and semantic goal understanding.

Contribution

The paper proposes a novel decoupled learning framework that separates perception from policy training, utilizing a self-supervised distance predictor for improved navigation performance.

Findings

VLD outperforms prior temporal distance methods like ViNT and VIP.

The approach achieves strong sim-to-real transfer in robotic navigation tasks.

Decoupled training enables scalable and robust goal-conditioned policies.

Abstract

Training end-to-end policies from image data to directly predict navigation actions for robotic systems has proven inherently difficult. Existing approaches often suffer from either the sim-to-real gap during policy transfer or a limited amount of training data with action labels. To address this problem, we introduce Vision-Language Distance (VLD) learning, a scalable framework for goal-conditioned navigation that decouples perception learning from policy learning. Instead of relying on raw sensory inputs during policy training, we first train a self-supervised distance-to-goal predictor on internet-scale video data. This predictor generalizes across both image- and text-based goals, providing a distance signal that can be minimized by a reinforcement learning (RL) policy. The RL policy can be trained entirely in simulation using privileged geometric distance signals, with injected noise to mimic the uncertainty of the trained distance predictor. At deployment, the policy consumes VLD predictions, inheriting semantic goal information-"where to go"-from large-scale visual training while retaining the robust low-level navigation behaviors learned in simulation. We propose using ordinal consistency to assess distance functions directly and demonstrate that VLD outperforms prior temporal distance approaches, such as ViNT and VIP. Experiments show that our decoupled design achieves competitive navigation performance in simulation with strong sim-to-real transfer, providing an alternative and, most importantly, scalable path toward reliable, multimodal navigation policies.