Temporal Straightening for Latent Planning

TL;DR

This paper introduces temporal straightening, a method that improves latent representations for planning by encouraging straight trajectories, leading to more stable gradient-based planning and higher success rates in goal-reaching tasks.

Contribution

It proposes a novel temporal straightening technique with a curvature regularizer to enhance latent planning representations, improving planning stability and success.

Findings

Latent trajectories become more straight, better approximating geodesic distances.

Planning becomes more stable with reduced curvature in latent space.

Success rates in goal-reaching tasks significantly increase.

Abstract

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.