Visual control improves with time-aware representation learning
Directed Temporal Representations for Offline Visual Control
Machine Learning
Summary
Controlling robots or agents by showing them images is hard because they need to understand how actions lead to changes over time. The authors propose a new way to teach machines to understand time-based relationships in visual scenes using offline videos, so the agent knows how close it is to a goal in terms of steps rather than just visual similarity. They build a system that estimates how long it takes to reach a goal and uses that to train the agent directly. Testing shows their approach helps agents find goals effectively without slow trial-and-error at test time.
What this means in practice
- •For robotic system developers: Train robots to estimate time-to-goal from video data, enabling efficient goal-directed control without online planning.
- •For autonomous vehicle engineers: Use temporal distance learning to improve decision-making based on visual inputs and predicted time costs to reach targets.
Authors
Chenyang Yuan, Haoyu Wang, Zhuo Sun, Xiaoyuan Cheng
Abstract
Predictive world models provide compact visual representations for control. Control requires a latent geometry aligned with temporal reachability rather than predictive similarity alone. We introduce Directed Temporal Representations for Control (DTRC), which learns such a geometry from offline visual trajectories on top of frozen LeWorldModel (LeWM) features. DTRC constructs a directed temporal quasimetric over the learned control representation. Short-range temporal offsets calibrate the distance scale. Bootstrapped targets extend temporal reachability across longer horizons. Action-conditioned consistency aligns the representation with local transition dynamics. The resulting distance estimates temporal reaching cost, and its change across a transition defines goal-relative temporal progress. We use this progress signal as a temporal critic for direct goal-conditioned policy learning. Model-assisted targets provide an additional training-time refinement under behavior-support and dynamics-agreement constraints. Across ten visual control tasks, DTRC achieves strong goal-conditioned control performance relative to planning and direct-policy baselines. Held-out diagnostics on the four LeWM tasks show consistent short-range temporal calibration, task-dependent long-range and directional structure, and positive transition-level progress. Temporal supervision improves the same flow-policy parameterization across all four LeWM tasks, while the resulting policy acts directly without iterative trajectory search at test time.