aiqbench analysis
This research introduces an adaptive rollout strategy for offline neural world models that terminates autoregressive predictions when epistemic uncertainty exceeds a specific threshold, significantly reducing computational costs without sacrificing accuracy. By leveraging ensemble methods and Monte Carlo Dropout with a stabilization warm-up phase, the approach achieves roughly 72% fewer cumulative rollout steps compared to fixed-horizon baselines while maintaining predictive reliability for model-based robotics.
Adaptive Rollout Truncation Based on Epistemic Uncertainty for Efficient Offline World Model Training
arXiv:2609.21482v1 Announce Type: new Abstract: Accurate neural world models are central to model-based robotics, where they enable robots to predict future states from previously observed trajectories. Multi-step autoregressive training improves long-horizon prediction, but fixed rollout horizons also increase computational cost and can amplify early training errors when the model is still inaccurate. Existing training schemes typically use the same rollout length throughout optimization, independent of the model's current predictive reliability. We propose an epistemic uncertainty-driven adaptive rollout strategy for offline world model training following an auto-curriculum training scheme. Instead of always unrolling to a fixed horizon, the model terminates autoregressive rollouts once epistemic uncertainty exceeds a threshold calibrated from a warm-up phase. We study two uncertainty estimators: a five-head ensemble with a shared recurrent backbone and Monte Carlo Dropout. A two-stage warm-up procedure stabilizes uncertainty estimates before we enable adaptive truncation. Experiments on ANYmal-D and ANT show that ensemble-based adaptive truncation matches or improves the prediction accuracy of fixed-horizon training and the RWM-U baseline while requiring substantially fewer cumulative rollout steps. Training a world model on ANYmal-D following the presented approach reaches comparable final performance with the baselines with roughly 72% less rollout computation. These results indicate that epistemic uncertainty is useful not only for downstream policy regularization, but also for making world model training itself more compute-efficient.
Abstract:Accurate neural world models are central to model-based robotics, where they enable robots to predict future states from previously observed trajectories. Multi-step autoregressive training improves long-horizon prediction, but fixed rollout horizons also increase computational cost and can amplify early training errors when the model is still inaccurate. Existing training schemes typically use the same rollout length throughout optimization, independent of the model's current predictive reliability. We propose an epistemic uncertainty-driven adaptive rollout strategy for offline world model training following an auto-curriculum training scheme. Instead of always unrolling to a fixed horizon, the model terminates autoregressive rollouts once epistemic uncertainty exceeds a threshold calibrated from a warm-up phase. We study two uncertainty estimators: a five-head ensemble with a shared recurrent backbone and Monte Carlo Dropout. A two-stage warm-up procedure stabilizes uncertainty estimates before we enable adaptive truncation. Experiments on ANYmal-D and ANT show that ensemble-based adaptive truncation matches or improves the prediction accuracy of fixed-horizon training and the RWM-U baseline while requiring substantially fewer cumulative rollout steps. Training a world model on ANYmal-D following the presented approach reaches comparable final performance with the baselines with roughly 72% less rollout computation. These results indicate that epistemic uncertainty is useful not only for downstream policy regularization, but also for making world model training itself more compute-efficient.
8 pages, 12 figures. Accepted at the IEEE/RSJ IROS 2026 Workshop "Rethinking Uncertainty for Modern Robotics Paradigms"
Robotics (cs.RO); Machine Learning (cs.LG)
arXiv:2609.21482 [cs.RO]
(or arXiv:2609.21482v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.21482 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 326 words
Related coverage
In-Distribution Imagination for Model-Based Offline Reinforcement Learning
arXiv · Oct 1, 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · Oct 13, 2026
WOMBET: World Model-Based Experience Transfer for Robust and Sample-efficient Reinforcement Learning
arXiv · Jun 12, 2026
The Machines that Make the Machines
NVIDIA · Oct 7, 2026