AI★★★★arXiv · 2026-07-16
RoboTTT: Context Scaling for Robot Policies
RoboTTT is a novel robot model and training method that scales visuomotor context to 8K timesteps, significantly enhancing robot capabilities.
📌 Key points
- RoboTTT can handle up to 8K timesteps of context without increasing inference la
- Supports one-shot imitation learning and on-the-fly policy improvement
- Enhances robustness to perturbations
- Improves performance on multi-stage long-horizon tasks
本页为 gitzw.com 基于公开来源的 AI 中文解读，非原文转载。