# AFTER-WM preprint reports lower robot prediction errors after dynamics change

_Published Saturday, November 14, 2026 at 4:05 AM EST · Science, AI · Latest · Tier 2 — Notable_

Researchers report in a preprint that AFTER-WM, a model for predicting robot behavior, reduced simulated multistep prediction error by 16%-34% against the strongest comparison model. It adapts to changes such as an unknown payload by combining a fast correction that fades with a slow correction that persists.

The method adjusts the weights of pretrained experts without changing those experts. Across five simulated robot datasets and real UAV data, it had lower multistep mean error than all ten comparison models and the highest mean return in a joint-failure control task. Code and datasets are promised for release.

## Sources

- [cs.AI updates on arXiv.org](https://arxiv.org/abs/2611.15215)

---
Canonical: https://techandbusiness.org/newswire/4QYfOliSPVMqU-OvZ1-NoV
Published: 2026-11-14T09:05:56.338Z
Story chronology: 2026-11-14T05:00:00.000Z
Retrieved: 2026-11-15T00:01:47.917Z
Publisher: Tech & Business (techandbusiness.org)