Keep the Future, Drop the Rollout: RIFT for World Action Models

arXiv:2608.11521v1 Announce Type: cross Abstract: World action models (WAMs) condition robot actions on predicted futures, but iterative video rollout increases deployment latency. We ask whether action generation requires the evolving rollout trajectory or only its future representation. Across four WAMs on all 40 LIBERO tasks, paired closed-loop interventions show that masking or reassigning future-cache values changes execution and reduces success, indicating sensitivity to future values and their assigned positions. For Joint and Cosmos-2, however, replaying one fixed final-clean key/value (K/V) cache nearly preserves unmodified execution, with 1.7 to 1.9cm end-effector average displacement error and 97.9% to 98.2% success. This separates cache consumption from production: these models can reuse a fixed cache but still require iterative rollout to construct it. We therefore propose RIFT (Rollout-free Imagination via Future Tokens), which uses learned anticipation tokens to construct a complete future K/V cache in one backbone pass while retaining the original future-read interface. On LIBERO, RIFT achieves 98.8% success, close to rollout-based Joint, IDM, and LingBot-VA at 98.4% to 98.6%, while reducing action-chunk latency by 68.2% to 89.1%. On RoboTwin2.0, RIFT reaches 92.9/92.6% on clean/randomized scenes, the highest observed among the evaluated methods. These results support rollout-free future conditioning without iterative video generation at deployment.

Source: arXiv cs.AI | 2026-08-13