RAI Institute
About
About RAI Institute
The Robotics and AI Institute conducts research in intelligent machines, combining robot learning, perception, control, hardware and large-scale data. Founded by Boston Dynamics founder Marc Raibert, the organisation builds experimental platforms and algorithms intended to make robots more capable, adaptable and useful in complex physical environments.
Research teams work across locomotion, manipulation, cognitive AI and robot design, using shared engineering infrastructure to test ideas on real machines. The institute publishes technical work, releases selected software and collaborates with academic and industrial researchers. Its model supports long-horizon research without tying every programme to an immediate commercial product.
Research
- RAI Institute demonstrates AthenaZero vision-based robot juggling
2026-05-27 - Research Publication - AthenaZero uses onboard visual feedback rather than an external motion-capture system. The multi-fingered system transitions between multiple juggling patterns. The work forms part of RAI Institute research in advanced learning for control.
Source - ZEST paper transfers athletic skills to Boston Dynamics Atlas
2026-01-30 - Research Publication - ZEST trains motion-imitation policies entirely in simulation with moderate domain randomization. Physical electric Atlas trials include army crawling, breakdancing and video-derived scene interactions such as box climbing. Adaptive sampling emphasizes difficult motion segments, while an assistive-wrench curriculum supports training of long-horizon manoeuvres.
Source
Timeline
Events
- 2026-09-18 - Acquisition - SoftBank agrees to acquire RAI Institute
The Cambridge, Massachusetts institute was founded by Marc Raibert in 2022 and researches robot control, perception, manipulation, navigation and physical AI. Public reporting says the transaction is undergoing review by the Committee on Foreign Investment in the United States. The agreement does not yet transfer ownership; completion remains subject to the pending review and other closing conditions.
Source - 2026-05-27 - Research Publication - RAI Institute demonstrates AthenaZero vision-based robot juggling
AthenaZero uses onboard visual feedback rather than an external motion-capture system. The multi-fingered system transitions between multiple juggling patterns. The work forms part of RAI Institute research in advanced learning for control.
Source - 2026-01-30 - Research Publication - ZEST paper transfers athletic skills to Boston Dynamics Atlas
ZEST trains motion-imitation policies entirely in simulation with moderate domain randomization. Physical electric Atlas trials include army crawling, breakdancing and video-derived scene interactions such as box climbing. Adaptive sampling emphasizes difficult motion segments, while an assistive-wrench curriculum supports training of long-horizon manoeuvres.
Source - 2025-02-05 - Partnership - Boston Dynamics and RAI Institute Partner to Advance Humanoid Robots Through Reinforcement Learning
The teams aim to improve transfer of learned mobility policies from simulation to physical hardware. Planned manipulation work combines walking with interaction involving fixtures such as doors and levers. Full-body contact research includes dynamic running and heavy-object manipulation requiring coordinated arms and legs.
Source
Relations
Relationships
- Boston Dynamics
Atlas Reinforcement Learning Research Partner - Boston Dynamics and RAI Institute are developing shared reinforcement-learning methods for Atlas mobility and whole-body manipulation.