Robotics · 2026-11-13
AFTER: A World Model with Adaptive Fast--Slow Timescales for Error-Driven Retention
Haoyi Zhang, Qi Zhang, Yunfan Ren, Yi Bin, Lei Zhu, Heng Tao Shen
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
AFTER-WM is a world model for online dynamics adaptation under sudden, persistent, initially unknown changes. It uses predictive correction from one-step prediction errors plus dual-timescale fast/slow adaptation (with fading fast correction and persisting slow correction) and multiple pre-trained experts whose combination weights are adaptively reweighted without altering experts. On five simulated robot datasets and real UAV data, the authors report lower multi-step mean error and the highest mean return in a joint-failure control task.
Why it matters
The paper targets a core world-model problem in robotics: adapting to abrupt, persistent dynamics changes while remaining responsive, stable (distinguishing transient disturbances), and persistent (not reverting when no new observations arrive during rollout). The authors aim to reduce prediction error accumulation during open-loop rollouts that can otherwise mislead planning and control.
Method
- Predictive correction with a mixture of K pre-trained expert dynamics models; during test time, online memories change based on one-step prediction errors without gradient updates to experts.
- Dual-timescale adaptation driven by prediction errors: a fast state that fades and a slow state that persists, with learned networks updating both for responsiveness, stability, and persistence.
- Multiple internal models: a router produces routing memory/combination weights that are updated from errors, while the experts themselves are never altered.
Limitation
In simulation, one MBCD seed’s predictions diverge on Panda and UR5e.
Abstract (from arXiv)
In robotics and embodied AI, a core challenge for world models is adapting to sudden, persistent, yet initially unknown changes in robot dynamics during interaction. For instance, when a robotic arm grasps an object with an unknown payload, the robot dynamics change suddenly. If the world model fails to adapt, its prediction errors accumulate over the open-loop rollout and ultimately mislead the robot's actions. Existing world models, however, struggle to simultaneously satisfy three critical requirements for dynamics adaptation: responsiveness, reacting promptly to sudden changes to curb prediction error accumulation; stability, distinguishing persistent changes from transient disturbances to prevent incorrect updates; and persistence, retaining the adapted dynamics without new observations so that later predictions do not revert to the old dynamics. Inspired by predictive coding in the human brain with multi-timescale adaptation, we propose AFTER-WM, a world model with Adaptive Fast-slow Timescales for Error-driven Retention, whose prediction errors drive a fast correction that fades and a slow correction that persists, while dynamically adjusting the combination weights of pre-trained experts without altering them, to overcome the trade-offs among responsiveness, stability and persistence. On five simulated robot datasets and real UAV data, AFTER-WM has a lower multi-step mean error than all ten compared models and the highest mean return in a joint-failure control task. In simulation, its multistep mean error is 16%-34% below the strongest compared model's and is also the lowest 1-4 steps after a change and under transient disturbances alone, and its step-40 error is the lowest as well, reflecting responsiveness, stability and persistence. All code (method, baselines, experiments) and datasets will be released online.
Related papers
- Few-Shot Residual Adaptation of Pretrained Latent World Models
- Causal world models under physical parameter shifts
- Utility-WAM: Correcting Future-Induced Action Changes
- HiLo-WAM: Decoupling Long-Horizon Task Evolution and Local Dynamics in World Action Models
- RM-WAM: Reliable Memory-Augmented World Action Models for Long-Horizon Robot Manipulation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First announced by arXiv: 2026-11-13 · Indexed by Pub-AI: 2026-11-13 UTC
AFTER: A World Model with Adaptive Fast--Slow Timescales for Error-Driven Retention
First announced by arXiv: 2026-11-13 · Original record ↗
- arxiv
- 2611.15215
Explicit source versions (1)
- v1 · Date not supplied
- Preprint available
Source date: 2026-11-13 · First detected: 2026-11-13 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2611.15215v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv
In robotics and embodied AI, a core challenge for world models is adapting to sudden, persistent, yet initially unknown changes in robot dynamics during interaction. For instance, when a robotic arm grasps an object with an unknown payload, the robot dynamics change suddenly. If the world model fails to adapt, its prediction errors accumulate over the open-loop rollout and ultimately mislead the robot's actions. Existing world models, however, struggle to simultaneously satisfy three critical requirements for dynamics adaptation: responsiveness, reacting promptly to sudden changes to curb prediction error accumulation; stability, distinguishing persistent changes from transient disturbances to prevent incorrect updates; and persistence, retaining the adapted dynamics without new observations so that later predictions do not revert to the old dynamics. Inspired by predictive coding in the human brain with multi-timescale adaptation, we propose AFTER-WM, a world model with Adaptive Fast-slow Timescales for Error-driven Retention, whose prediction errors drive a fast correction that fades and a slow correction that persists, while dynamically adjusting the combination weights of pre-trained experts without altering them, to overcome the trade-offs among responsiveness, stability and persistence. On five simulated robot datasets and real UAV data, AFTER-WM has a lower multi-step mean error than all ten compared models and the highest mean return in a joint-failure control task. In simulation, its multistep mean error is 16%-34% below the strongest compared model's and is also the lowest 1-4 steps after a change and under transient disturbances alone, and its step-40 error is the lowest as well, reflecting responsiveness, stability and persistence. All code (method, baselines, experiments) and datasets will be released online.
