RoboTTT: Context Scaling for Robot Policies

2026-07-16

AI Summary

Existing robotic foundation models are constrained by limited visual-motor context lengths, hindering their ability to handle long-horizon, multi-stage tasks. This work proposes a test-time training mechanism that integrates fast weight recurrent states, sequence-level action forcing, and truncated backpropagation through time to extend the effective context length to 8K timesteps without increasing inference latency. For the first time, this approach demonstrates consistent closed-loop performance gains as a function of pretraining context length. Evaluated on real robots, the method achieves an 87% overall performance improvement, successfully completing a five-minute, ten-stage assembly task. The 8K-context model outperforms its 1K counterpart by 62% and enables one-shot imitation from human videos, online policy adaptation, and highly robust control.

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/

Problem

• robot foundation models
• visuomotor context
• long-horizon tasks
• context scaling
• in-context imitation

Innovation

• Test-Time Training
• Long-Context Modeling
• Fast Weights
• Robot Foundation Models
• Visuomotor Policy