Search papers, labs, and topics across Lattice.
This paper introduces temporal straightening, a curvature regularizer applied during world model training, to encourage locally straightened latent trajectories. By minimizing the curvature of latent space trajectories, the method aims to make Euclidean distance a better proxy for geodesic distance, thus improving the conditioning of the planning objective. Experiments demonstrate that this approach stabilizes gradient-based planning and achieves higher success rates in goal-reaching tasks.
Straightening latent space trajectories with a simple curvature regularizer dramatically improves the stability and success of gradient-based planning in world models.
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.
Meta AI