Covariant

Earned a Ph.D. in EECS at UC Berkeley under Pieter Abbeel, researching reinforcement learning and robotics at Berkeley's AI Lab and as a research scientist at OpenAI. Before Covariant, co-founded Sellegit, an earlier startup. At Covariant, led development of the Covariant Brain — an AI layer for industrial robots — and later the RFM-1 multimodal foundation model. The venture raised over $350M before Amazon acqui-hired Chen, Abbeel, and Rocky Duan along with roughly a quarter of staff in August 2024; Chen now heads Frontier AI & Robotics at Amazon.

A professor of EECS at UC Berkeley since 2008, Director of the Berkeley Robot Learning Lab, and Co-Director of the Berkeley AI Research (BAIR) Lab. He earned a BS/MS in Electrical Engineering from KU Leuven (2000) and a PhD in Computer Science from Stanford (2008) under Andrew Ng, with a dissertation on apprenticeship learning that famously enabled autonomous helicopter aerobatics beyond what experienced human pilots could execute. Before Covariant, he co-founded Gradescope (2014), an AI-assisted grading platform, and spent time as a Research Scientist at OpenAI. His Berkeley lab's work on deep reinforcement learning and robot manipulation directly seeded Covariant's approach. He also hosts *The Robot Brains* podcast. In August 2024, he was among the Covariant founders and ~25% of staff acqui-hired by Amazon.

Earned a PhD from UC Berkeley under Pieter Abbeel, with a dissertation on meta-learning and cloud robotics benchmarks. Spent 2016–2017 as a research scientist at OpenAI, where he worked on improving sample efficiency in robotic learning. Co-founded Covariant (originally Embodied Intelligence) in 2017 as its CTO, co-architecting the Covariant Brain — an AI platform enabling robots to pick and sort disparate items in warehouses — and later the RFM-1 multimodal foundation model. In August 2024, Amazon acqui-hired the founding team and roughly a quarter of staff; Duan now leads research at Amazon's Frontier AI & Robotics group. Named to Forbes 30 Under 30 in 2024.

Earned a double-major B.A. in Computer Science and Statistics from UC Berkeley, then entered the EECS PhD program under Pieter Abbeel at Berkeley's Robot Learning Lab. There, Zhang researched deep imitation learning and one-shot robot learning from human video demonstrations—work that directly fed into Covariant's approach to general-purpose robotic manipulation. Before founding Covariant in 2017, he also spent time at OpenAI alongside his future co-founders. When Amazon acqui-hired Pieter Abbeel, Peter Chen, and Rocky Duan in August 2024—taking roughly a quarter of the company—Zhang was the sole co-founder to stay. He now co-leads the remaining independent Covariant with Ted Stinson.