RFM-1

Architecture

Multimodal any-to-any autoregressive sequence model (~8B params); next-token prediction over text/images/video/robot-actions/sensors.

Key facts

Model

RFM-1 (Robotics Foundation Model), announced Mar 2024 at MODEX: ~8B-param multimodal any-to-any sequence model (text/images/video/robot-actions/sensors). Proprietary.

Deployed wiring (verified)

Pre-acquisition, the Covariant Brain ran on real warehouse pick-and-place arms at NAMED customers: Otto Group, Radial, McKesson. Genuine deployed wiring, not a demo.

Corporate state (Amazon absorption)

Aug 30 2024 reverse-acquihire: Amazon hired founders Abbeel/Chen/Duan + ~25% staff + took a NON-EXCLUSIVE license to Covariant's models ($380M + $20M per FTC whistleblower). Independent Covariant winding down; Amazon is LICENSEE, not the brain developer.

Cap-flag

No peer-reviewed paper, no model card, no released checkpoints (only a company tech blog). 'Human-like reasoning' is marketing. Current independent 2026 deployments + whether RFM-1 (vs the older Covariant Brain) runs inside Amazon are unverified.

Developed by (1)

• Covariant

Common questions

What is RFM-1?

Covariant (founded 2017 in the Berkeley area by Pieter Abbeel, Peter Chen, Rocky Duan, and Tianhao Zhang) built RFM-1, the Robotics Foundation Model announced in March 2024: a roughly 8-billion-parameter multimodal any-to-any sequence model trained on text, images, video, robot actions, and sensor readings, used to power warehouse pick-and-place robots. The registry records the brain at commercial maturity to reflect genuine deployed wiring: pre-acquisition, the Covariant Brain ran on real robotic arms at named customers including Otto Group, Radial, and McKesson, not just in demonstrations. Its corporate state is a key big-tech-absorption datapoint: in a reverse-acquihire that closed on August 30, 2024, Amazon hired founders Abbeel, Chen, and Duan along with about a quarter of staff and took a non-exclusive license to Covariant's models, reportedly for $380 million plus a $20 million payment per an FTC whistleblower complaint, with independent Covariant winding down and Amazon as a licensee rather than the brain developer. RFM-1 is proprietary with no peer-reviewed paper, model card, or released checkpoints, the human-like-reasoning framing is marketing, and current independent deployments and whether RFM-1 itself runs inside Amazon are not verified.

What type of AI is RFM-1?

RFM-1 is a foundation model on the DEPLOY registry. It is proprietary.

Who developed RFM-1?

RFM-1 is credited to Covariant on the DEPLOY registry. Each developer attribution is verified via primary sources.

Which robots run on RFM-1?

No robot models on the DEPLOY registry are recorded as running RFM-1. DEPLOY wires brain-to-model connections only when the wiring is verifiable from primary sources; absence may reflect pre-deployment or unverified manufacturer claims.

Is RFM-1 open source?

No. RFM-1 is recorded as proprietary on the DEPLOY registry. Model weights and source are not publicly available.

What is RFM-1's maturity stage?

RFM-1 is at the commercial stage on the DEPLOY maturity ladder. Commercial stage means production-grade deployments are operating at named customer sites.

Sources (4)

1. https://www.businesswire.com/news/home/20240311948570/en/Covariant-Introduces-RFM-1-to-Give-Robots-the-Human-like-Ability-to-Reason ↗
2. https://www.aboutamazon.com/news/company-news/amazon-covariant-ai-robots ↗
3. https://www.geekwire.com/2024/amazon-hires-covariant-founders-inks-licensing-deal-with-robotics-ai-startup-in-latest-reverse-acquihire-deal/ ↗
4. https://www.washingtonpost.com/technology/2025/01/18/amazon-antitrust-ai-whistleblower/ ↗