Parallel predictive models speed up and improve long-term planning
Parallel Predictive World Models for Accurate and Efficient Long-Horizon Planning
Artificial Intelligence
Summary
Planning ahead in complex environments usually requires predicting one step at a time, which is slow and can compound errors. The authors introduce Parallel Predictive World Models (PPWM), which predict entire sequences of future events at once while keeping important causal relationships. This method reduces errors and runs more than three times faster than traditional step-by-step predictions. Their tests on visual control tasks show that PPWM is both more accurate and efficient for long-term planning.
What this means in practice
- •For robotics engineers: Improve robot control systems by enabling faster and more accurate long-term planning in simulated environments.
- •For video game developers: Boost the speed and accuracy of AI decision-making for complex in-game planning over long sequences of actions.
Authors
Wanjin Feng, Baobin Zhang, Ao Yu, Shibo Feng, Xi Wang, Xingyu Gao
Abstract
Long-horizon world-model planning typically relies on autoregressive rollouts, where predicted states are repeatedly fed back into the model. This preserves temporal structure but creates a horizon-length sequential path and exposes later predictions to recursive decoded-state feedback. We introduce Parallel Predictive World Models (PPWM), which predict a finite-horizon trajectory in parallel while retaining causal interaction among future representations. Each horizon is conditioned on its causal action prefix, and future representations interact before decoding, separating temporal causality from state-by-state output recursion. We formalize this distinction by viewing autoregressive rollout as a causal trajectory map and identifying the decoded-state feedback pathway removed by PPWM. Across four visual-control tasks, PPWM achieves the lowest long-horizon prediction error and the highest Cross-Entropy Method (CEM) simulator success among the evaluated predictive interfaces. Meanwhile, PPWM achieves more than a 3$\times$ average CEM planning speedup over the autoregressive LeWM baseline. These results suggest that accurate and efficient long-horizon world-model planning does not require state-by-state autoregression, but can instead be achieved through parallel causal trajectory prediction.