TTT in VLA 模型：探索用人类示范进行训练
TTT in VLA Models: Exploring Training with Human Demonstrations
AI 导读
一项机器人研究将 Test-Time Training（TTT，用快权重存储上下文）集成到 InternVLA-A1 和 Pi0.5 中，通过成对人类示范与机器人轨迹训练研究动作生成。实验在视频、短视频和单帧线索任务上考察了训练目标、梯度覆盖与对比正则化。结果显示，在当前规模下加入 TTT 及相关设置未能实现基于示范的一致目标选择，作者公开实现与配置供复现。
正文 · 原文
Abstract:Long-horizon robot tasks require retaining and using past information. We integrate Test-Time Training (TTT), which stores context in fast weights, into InternVLA-A1 and Pi0.5 to study action generation conditioned on human demonstrations through training on paired demonstrations and robot trajectories. We implement memory transfer between the two phases and truncated backpropagation, and examine training objectives, gradient coverage, and contrastive regularization across video, short-video, and single-frame cue tasks. In our experiments, adding TTT and training on paired data at the current scale did not lead to consistent target selection based on demonstrations. Under the tested settings, shorter sequences, full trajectory gradient connectivity, and contrastive regularization did not clearly improve responsiveness to demonstrations. When demonstrations were reduced to single-frame cues, real-robot execution still showed a preference for cups at fixed positions. We document the implementation, training configurations, and observations to support further inspection and replication.
| Comments: | 22 pages, 6 figures, a techreport form standard-robots company |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2610.32094 [cs.RO] |
| (or arXiv:2610.32094v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.32094 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Qiangong Zhou [view email]
[v1]
Fri, 23 Oct 2026 02:55:23 UTC (2,483 KB)
来源：arXiv cs.AI · arxiv.org