Pieter Abbeel

Who they are

Pieter Abbeel is a Distinguished Scientist at Amazon — co-founded Covariant (AI warehouse robotics), Gradescope (acquired), and AIX Ventures, and hosts The Robot Brains podcast.

Person

Pieter Abbeel started in Belgium — BS/MS in Electrical Engineering at KU Leuven — before crossing to Stanford for a PhD in AI in 2008, advised by Andrew Ng. From Stanford he moved to UC Berkeley as a professor of Electrical Engineering and Computer Sciences, where he ran the Berkeley Robot Learning Lab and became one of the field's central figures in deep reinforcement and imitation learning. He did a stint at OpenAI as a research scientist before co-founding Covariant in 2017 with students Peter Chen, Rocky Duan, and Tianhao Zhang — building the Covariant Brain, an AI platform for warehouse robot automation. In August 2024 Amazon brought him in as an Amazon Scholar to co-lead its Frontier AI & Robotics team, part of a reverse acquihire that also pulled in his co-founders. Beyond Covariant, he co-founded Gradescope (AI-powered grading, now acquired), SafelyYou (AI safety for seniors), AIX Ventures (a VC fund backing AI startups, launched 2021), XDOF (robot foundation model infrastructure), and Engram (AI that builds compact user-specific memories). He hosts The Robot Brains podcast, has spoken at NIPS keynotes in 2016 and 2017, appeared on the Lex Fridman MIT AI Podcast on deep reinforcement learning, and has given C-suite briefings to Microsoft, Google, McKinsey, and others. The through-line: he keeps one foot in research and one in company-building — every move expands the surface area between academic robotics and deployed commercial systems.

Company

The most recent development at Covariant is the 2026 announcement of RFM-1 — a Robotics Foundation Model with 8 billion parameters, multimodal, designed to give robots human-like reasoning about the physical world. This followed the May 2025 acquihire deal in which Amazon hired all three founders (Abbeel, Peter Chen, Rocky Duan) and roughly 25% of Covariant's employees, with the deal valued at $380M plus a $20M licensing payment for Covariant's robotic foundation models on a non-exclusive basis. Post-acquihire, Ted Stinson now serves as CEO of the remaining Covariant entity alongside co-founder Zhang Tianhao. Before the acquihire, Covariant had raised $222M across five rounds including an $80M Series C in April 2023, with customers including logistics provider Radial and German e-commerce giant Otto Group.

Market

Covariant operates in AI-driven warehouse automation, a market growing rapidly as retailers and logistics operators respond to labor shortages — with over 80% of retail leaders citing automation as a strategic priority. Its main competitors include Boston Dynamics, Symbotic, Berkshire Grey, RightHand Robotics, Universal Robots, and Ocado Technology, across an active field of more than 100 competitors. US-China technological rivalry and regulatory fragmentation in AI governance add geopolitical complexity for any company in this space operating across multiple countries.

Network

Abbeel's closest professional ties are his Covariant co-founders — Peter Chen, Rocky Duan, and Tianhao Zhang — who moved with him into Amazon's robotics division. On the investor and advisory side he has board-level relationships with Mike Volpi at Index Ventures and Jordan Jacobs at Radical Ventures, and sits on the advisory board of Reflection AI. He is also an investor and co-founder at Engram, and an investor at autonomous driving company Waabi.

Peter Chen· Co-founder, Covariant
Rocky Duan· Co-founder, Covariant
Tianhao Zhang· Co-founder, Covariant
Andrew Ng· PhD advisor, Stanford University
Mike Volpi· Partner, Index Ventures
Jordan Jacobs· Co-founder, Radical Ventures

How they likely show up

Parallel track of professor + serial founder (Covariant, Gradescope, SafelyYou, AIX Ventures, XDOF, Engram) → extremely high agency; he doesn't wait for permission to start something.
Moved from academic lab (Berkeley) → OpenAI → own company → Amazon Scholar role → suggests he tests ideas at institutional scale before backing them with a company, not the reverse.
Hosts The Robot Brains podcast and has given keynotes at NIPS 2016 and 2017 → comfortable as a public explainer; will likely engage more deeply if you engage on substance rather than titles.
Role type classified as 'hybrid' (researcher + executive + investor) → probably operates across multiple registers in a single conversation; don't pigeonhole him as only a scientist or only a founder.
C-suite speaking engagements for Microsoft, Google, McKinsey, Coca-Cola → accustomed to translating deep technical work for non-technical audiences; won't talk down to you but will calibrate depth to your cues.
Multilingual (English, Dutch, French) and Belgian-educated before US career → international lens baked in early; comfortable operating across cultural and institutional contexts.

Conversation tips

→ Ask about RFM-1 specifically — what an 8-billion-parameter multimodal robotics model changes about how robots generalize versus task-specific models. He'll have strong views.
→ Reference The Robot Brains podcast by name and a specific episode theme (embodied intelligence, meta-learning) — it signals you know his public intellectual interests, not just his CV.
→ He's been a founder, a professor, a VC (AIX Ventures), and now an Amazon Scholar — ask which lens he finds himself using most at Amazon, and you'll learn how he's orienting his current chapter.
→ If you work in logistics or warehouse automation, come with a specific operational problem — he's done customer deployments with Radial and Otto Group and will engage concretely, not hypothetically.
→ Don't treat him purely as an Amazon employee — Covariant, Gradescope, and his portfolio companies are a bigger part of his identity than any single employer.

Toolbox

Openers

Open on RFM-1 — Covariant announced an 8-billion-parameter multimodal robotics foundation model in early 2026, and the question of whether foundation model approaches from language will transfer to physical robot manipulation is genuinely unsettled. He has skin in that answer.
Reference the Amazon acquihire structure — the deal was explicitly structured as a reverse acquihire (founders hired, technology licensed non-exclusively) to sidestep antitrust scrutiny. That's a legally and strategically unusual move, and he lived through the design of it.
Bring up Gradescope — he co-founded it in 2014 as an AI grading tool for educators, and it was acquired. It's an easy-to-overlook company in his portfolio given the robotics focus, but it shows he was building applied AI products a decade before it was fashionable.

Discovery questions

With RFM-1 now out, how do you think about the difference between a robotics foundation model and a language model fine-tuned for robot control — is it fundamentally a different training problem or a data problem?
When you moved from running the Berkeley Robot Learning Lab to co-founding Covariant, what was the thing that academic research couldn't give you — and has Amazon now given you a version of that at scale?
AIX Ventures launched in 2021 to back AI startups — what do you look for in an AI robotics founder that you wouldn't have known to look for before building Covariant?

Avoid

Don't open with broad questions about 'the future of AI' — he's a keynote speaker who fields those constantly; go specific on a product, a model, or a company decision and he'll engage at a different level.

Make it yours