Temporal Straightening for Latent Planning
This work addresses the challenge of improving planning efficiency in AI systems, though it is incremental as it builds on existing world model frameworks.
The paper tackled the problem of learning representations for latent planning by introducing temporal straightening to reduce irrelevant information in visual features, resulting in more stable gradient-based planning and significantly higher success rates in goal-reaching tasks.
Learning good representations is essential for latent planning with world models. While pretrained visual encoders produce strong semantic visual features, they are not tailored to planning and contain information irrelevant -- or even detrimental -- to planning. Inspired by the perceptual straightening hypothesis in human visual processing, we introduce temporal straightening to improve representation learning for latent planning. Using a curvature regularizer that encourages locally straightened latent trajectories, we jointly learn an encoder and a predictor. We show that reducing curvature this way makes the Euclidean distance in latent space a better proxy for the geodesic distance and improves the conditioning of the planning objective. We demonstrate empirically that temporal straightening makes gradient-based planning more stable and yields significantly higher success rates across a suite of goal-reaching tasks.