MoMo: Conditioned Contrastive Representation Learning for Preference-Modulated Planning

TL;DR

MoMo introduces a preference-conditioned contrastive planning method that allows real-time modulation of plan conservativeness, improving safety and consistency without retraining.

Contribution

It presents a novel contrastive planning approach that incorporates user preferences through feature-wise modulation, enabling dynamic safety adjustments at inference.

Findings

MoMo adapts plan safety smoothly according to user preferences.

It maintains inference efficiency by preserving probability density ratios.

Outperforms state augmentation baselines across six environments.

Abstract

Temporally contrastive representation learning induces a latent structure capable of reducing long-horizon planning to inference in a low-dimensional linear system. However, existing contrastive planning work learns a single latent geometry which cannot distinguish multiple valid behaviors trading task efficiency against risk exposure for the same start-goal query. We introduce MoMo, a preference-conditioned contrastive planner allowing a scalar user preference to continuously modulate plan conservativeness at inference time, without retraining. MoMo learns a joint conditioning of the representation geometry and latent prediction operator via Feature-Wise Linear Modulation and low-rank neural modulation, respectively. We show that our formulation preserves the probability density ratio encoded in the representation space that is required for inference-driven contrastive planning, further retaining its inference-time efficiency. Across six environments, MoMo smoothly adapts plan safety according to user preferences, yielding improved temporal and preferential consistency over state augmentation baselines.