RoboTTT: Context Scaling for Robot Policies

RoboTTT: Context Scaling for Robot Policies

|Authors: Yunfan Jiang, Yevgen Chebotar, Ruijie Zheng, Fengyuan Hu, Yunhao Ge, Jimmy Wu, Tianyuan Dai, Scott Reed, Li Fei-Fei, Yuke Zhu, Linxi “Jim” Fan

|Venue: arXiv preprint | Year: 2026

|arXiv: 2607.15275v1

Abstract

Recent robot foundation models operate with single-step or short-history visuomotor context. We introduce Test-Time-Training Robot Policies (RoboTTT), a robot model and training recipe that scale visuomotor context to 8K timesteps, three orders of magnitude beyond state-of-the-art policies, without growing inference latency. At this context length, we unlock new robot capabilities: one-shot in-context imitation from human video demonstrations, on-the-fly policy improvement, robustness to perturbations, and stronger performance on multi-stage, long-horizon tasks. We also observe, for the first time, steady gains in closed-loop performance as pretraining context length scales. At its core, RoboTTT integrates Test-Time Training into robot foundation models such as Vision-Language-Action policies, yielding a sequence model whose recurrent state consists of fast weights, parameters updated by gradient descent during both training and inference, compressing histories into weight space and retrieving contextual information for long-context conditioning. To scale training context length, the recipe combines sequence action forcing with truncated backpropagation through time. On challenging real-robot manipulation tasks, RoboTTT improves overall performance by 87% over the single-step context baseline and fully completes a five-minute, ten-stage assembly task, which no baseline ever does. RoboTTT trained with 8K-timestep context outperforms the same model pretrained with 1K timesteps by 62%, suggesting context length as a new scaling axis for robot foundation models. Videos are available at https://research.nvidia.com/labs/gear/robottt/

Key Contributions

• Recent robot foundation models operate with single-step or short-history visuomotor context
• We introduce Test-Time-Training Robot Policies (RoboTTT), a robot model and training recipe that …
• At this context length, we unlock new robot capabilities: one-shot in-context imitation from huma…
• We also observe, for the first time, steady gains in closed-loop performance as pretraining conte…
• At its core, RoboTTT integrates Test-Time Training into robot foundation models such as Vision-La…

Topics

• manipulation
• vision
• vla
• reinforcement-learning
• learning-from-demonstration

Code & Data

• GitHub repository: https://github.com/fla-org/flash-linear-attention

BibTeX

@article{Jiang2026_260715275v1,
  title     = {RoboTTT: Context Scaling for Robot Policies},
  author    = {Yunfan Jiang and Yevgen Chebotar and Ruijie Zheng and Fengyuan Hu and Yunhao Ge and Jimmy Wu and Tianyuan Dai and Scott Reed and Li Fei-Fei and Yuke Zhu and Linxi "Jim" Fan},
  year      = {2026},
  eprint    = {2607.15275v1},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  url       = {https://arxiv.org/abs/2607.15275v1}
}