REKI: Registering Knowledge in VLAs

Key points

• REKI consists of a vision-language model (VLM) followed by a flow-matching action expert, and is pretrained on a curated mixture of public robot, egocentric human and vision-language data.1
• Additionally, we endow it with a memory that gives the policy access to its recent past at negligible cost: a few register tokens in the action expert summarise the output of the VLM at each frame and persist in the expert's cache.1
• Kyutai, with Inria and École des Ponts, has released REKI, a 2.4B vision-language-action model (VLA) for robot manipulation.2
• The REKI vision-language-action model is pretrained only on public data.2

Sources (2)

• [1]REKI: Registering Knowledge in VLAs
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 04:34 PM

  We introduce REKI, a 2.4B vision-language-action model (VLA) pretrained exclusively on public data that sets a new Pareto frontier between performance and model size: on LIBERO-Plus, RoboTwin Clean2Rand and RoboCasa365, it is outperformed only by models about twice its size.

  REKI consists of a vision-language model (VLM) followed by a flow-matching action expert, and is pretrained on a curated mixture of public robot, egocentric human and vision-language data.

• [2]Meet Kyutai REKI: The 2.4B VLA Model Bringing Efficient Memory to Robot Manipulation
  MarkTechPost · Oct 17, 08:22 PM

  Kyutai, with Inria and École des Ponts, has released REKI, a 2.4B vision-language-action model (VLA) for robot manipulation.

  The REKI vision-language-action model is pretrained only on public data.

Extractive summary: sentences quoted from the sources.