Summary
Jason Ma (Yecheng Ma) is a co-founder and Chief Scientist of Dyna Robotics, which builds general-purpose, dual-arm robots powered by a proprietary embodied AI foundation model. He brought deep robot-learning research expertise to the company after prior research roles at Google DeepMind, NVIDIA AI, and Meta AI, and a Ph.D. from the University of Pennsylvania’s GRASP Laboratory.
Career
Ma earned an undergraduate degree in Computer Science from Harvard University before pursuing a Ph.D. in Computer and Information Science at the University of Pennsylvania’s GRASP Laboratory, where he was advised by Dinesh Jayaraman and Osbert Bastani and worked within the Perception, Action, and Learning research group. His doctoral research focused on reinforcement learning and robot learning, and in particular on training and deploying foundation models that let robots learn from internet-scale, multi-modal data rather than requiring task-specific engineering (GRASP Lab).
During his Ph.D., Ma was named an Apple Scholar in AIML, a fellowship recognizing his contributions to computer science and engineering, and received an OpenAI Fellowship and an RSS Pioneers award. He was a lead author on NVIDIA’s Eureka project, which used large language models to automate reward design for reinforcement learning, work that was subsequently included among NVIDIA Research’s top 10 projects of 2023. His broader publication record spans robot manipulation skill transfer from web video (ZeroMimic), vision-language modeling of articulated objects (Articulate-Anything), and safe model-based reinforcement learning, and a paper of his was a Best Paper Finalist at ICRA 2024. Alongside his doctoral work, he held research positions at Google DeepMind, NVIDIA AI, and Meta AI (GRASP Lab — Apple Scholar announcement; Jason Ma personal website).
Ma completed his Ph.D. in 2025 and co-founded Dyna Robotics with repeat entrepreneurs Lindon Gao and York Yang, serving as the company’s Chief Scientist. According to Dyna, Gao and Yang — who had previously built and sold smart-cart startup Caper AI to Instacart for $350 million — sought out Ma specifically to pair their experience shipping production-grade AI hardware with deep robotics research expertise. At Dyna, Ma has led development of the company’s DYNA-1 foundation model and its research into foundation reward models for scalable robot learning, a topic he has since presented publicly at industry venues such as Foxglove’s Actuate robotics developer conference (The Robot Report; Actuate 2025).
Source Notes
This profile draws on Jason Ma’s University of Pennsylvania GRASP Laboratory biography and Apple Scholar announcement, his personal academic website and publication list, a Foxglove Actuate 2025 conference speaker profile, and The Robot Report’s coverage of Dyna Robotics’ Series A funding round. His exact Ph.D. completion date is approximate, based on his GRASP Lab listing of “PhD, CIS ‘25.”