The RAI Institute (formally the Boston Dynamics Ai Institute)

What The RAI Institute (formally the Boston Dynamics Ai Institute) does

The RAI Institute (Robotics and AI Institute) is a private, non-commercial research institute incorporated as Robotics and AI Institute LLC in Massachusetts. Founded in 2022 by Marc Raibert — the founder of Boston Dynamics — and rebranded to its current name in 2024, the institute operates from a 251–500-person team across offices in Cambridge, MA (145 Broadway) and Zürich, Switzerland.

Its research is organized into five vertical programs: Dexterous Manipulation, Advanced Learning for Control, Data-Driven AI Models for Physical Robot Interaction, Navigation in Challenging Environments, and Ethics of Robots in Society. The institute builds proprietary robotic platforms (AthenaZero multi-fingered juggling hand, UMV ultra-mobility vehicle, Koala co-designed gripper system), develops novel ML algorithms (Q2RL, Show Don't Tell, LIDE, ExpertGen, Real-is-Sim, Diffuse-CLoC, AnyTask), and pairs each hardware platform with purpose-built data-collection methodologies — most notably the handheld gripper that closes the embodiment gap between human demonstration and robot execution.

The institute does not sell products, charge for access, or hold a commercial go-to-market motion. Revenue is not disclosed; outputs are released as open-source software on GitHub, open-access papers (arXiv and the institute's resources page), and video demonstrations via YouTube, LinkedIn, X, and Instagram. Audience and 'customers' are the academic and industrial robotics research community, conference attendees at ICRA/HRI/RSS/CHI/IDC, and the visiting-fellows ecosystem. Revenue mechanics are therefore grant-, sponsor-, or donor-funded, not transactional.

Name

The RAI Institute (formally the Boston Dynamics Ai Institute)

Legal name

Robotics and AI Institute LLC

Website

https://rai-inst.com

Company type

Private

Founded year

2022

Operating status

Operating

Headcount range

251–500 employees

Short description

The RAI Institute is a private, non-commercial robotics and AI research organization founded in 2022 by Marc Raibert, operating from Cambridge, MA and Zürich. It conducts open research across dexterous manipulation, robot learning, navigation, and robot ethics, releasing hardware platforms, algorithms, and open-source software to the global robotics research community.

Ownership category

akta.pro rank

Product category

Robotics and AI Research

NAICS

Research and Development in the Physical, Engineering, and Life Sciences (except Nanotechnology and Biotechnology) (541715)

SIC

Services-Commercial Physical & Biological Research (8731)

akta.pro primary industry

Robot Software, Autonomy Stacks & Simulation (ROS, digital twins) (HDAAAIAK)

Keywords

• Robotics research
• Artificial intelligence
• Intelligent machines
• Robot manipulation
• Reinforcement learning

Headquarters

HQ city

Cambridge

HQ country

United States

HQ region

North America

Offices3 records

Markets served

GTM type

B2B

Offering type

Services

Cost components

Personnel, Technology or R&D, Infrastructure, Operations

Distribution channels1 record

Marketing channels9 records

Core offering

The RAI Institute conducts foundational research to develop future generations of intelligent robots across five areas: Dexterous Manipulation, Advanced Learning for Control, Data-Driven AI Models for Physical Robot Interaction, Navigation in Challenging Environments, and Ethics of Robots in Society. The institute publishes open-access research papers, releases open-source software packages (judo, AnyTask, ExpertGen, Q2RL, Diffuse-CLoC, Real-is-Sim, ReLIC), and develops physical robot platforms (AthenaZero juggling hand, Ultra Mobility Vehicle, Koala gripper system) shared with the global research community. As a research institute, the organization does not sell commercial products; all outputs are made publicly available.

Product overview

The RAI Institute (Robotics and AI Institute) is a research institute focused on developing future generations of intelligent machines. Their research portfolio includes physical robot platforms (UMV, AthenaZero, Koala gripper system), open-source software packages (judo, AnyTask, ExpertGen, Q2RL, Diffuse-CLoC, Real-is-Sim), and research methodologies (Show Don't Tell, ReLIC). These tools span five core research areas: Dexterous Manipulation, Advanced Learning for Control, Data-Driven AI Models for Physical Robot Interaction, Navigation in Challenging Environments, and Ethics of Robots in Society. The institute operates as a research organization without commercial product offerings, sharing their work through open-source code releases, research papers, and public demonstrations.

Differentiator

Problem solved

Functional benefit

Products and services

• judo An open-source software package for sampling-based model predictive control, providing robust implementations of common algorithms, a comprehensive suite of benchmark tasks, and a simple but extensible interface over a MuJoCo physics backend. Intended for robotics researchers and practitioners working on model predictive control.
• AnyTask A pipeline for simulation data generation that leverages large language models to generate task descriptions, code for task success checkers, scripted policies, and reward functions for automated data generation in robotics research.
• ExpertGen A pipeline for generating policies in simulation that combines semantic priors of scripted policies with the reactivity and success rates of reinforcement learning, using diffusion priors and massively parallel RL to optimize contact-rich manipulation policies.
• Q2RL An open-source algorithm (Q-Estimation and Q-Gating from BC for RL) for efficient offline-to-online learning that extracts Q-functions from behavior cloning policies and uses Q-gating to switch between BC and RL actions, achieving up to 100% success rates on manipulation tasks.
• Diffuse-CLoC An open-source guided diffusion framework that bridges kinematic motion diffusion models with physics-based control policies for physics-based character look-ahead control, training end-to-end state-action diffusion models without retraining.
• Real-is-Sim An open-source framework that integrates simulation into behavior cloning pipelines using a dynamic digital twin powered by the Embodied Gaussian simulator that synchronizes with the real world at 60Hz, driving policy execution while maintaining real-time synchronization.
• ReLIC Framework An open-source reinforcement learning framework for flexible loco-manipulation that enables robots to dynamically coordinate arms, legs, and body movements for complex tasks such as tire uprighting, rolling, dragging, and stacking.

Quantifiable outcome

• AthenaZero can learn key skills such as cascade juggling in under 10 minutes of real interaction
• +3 more outcomes

Segments1 record

Ideal customer profiles1 record

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability24 records

Feature10 records

Scale indicators1 record

Recent moves6 records

Expansion highlights6 records

Emerging players

• 1X Technologies: Builds the NEO humanoid robot with a strong AI/foundation-model research group. Comparable to RAI's whole-body manipulation and learning-for-control research themes, though shipping a consumer product rather than operating as an open research institute.
• Agility Robotics: Creator of the Digit humanoid robot, with substantial research effort on bipedal locomotion and manipulation. Comparable to RAI's navigation-in-challenging-environments and dynamic whole-body manipulation themes, but commercially productized.
• Physical Intelligence: A well-funded startup building foundation models for general-purpose robot control and manipulation. Highly comparable in research agenda (foundation models for physical robot interaction, manipulation) but operating with a commercial rather than open-IP research model.
• Skild AI: An emerging robotics AI startup developing general-purpose "robot brain" foundation models. Comparable to RAI's data-driven AI models research area as a commercial competitor chasing the same embodied foundation-model opportunity.
• Figure AI: Developer of the Figure humanoid robot and Helix foundation model. Comparable to RAI through overlapping navigation-in-challenging-environments and whole-body manipulation research, but with a humanoid-product commercial focus.

Broad incumbents

• Google DeepMind (Robotics): DeepMind's Robotics and embodied AI teams (RT-2, AutoRT, etc.) pursue foundation models for physical robot interaction — directly overlapping RAI's dexterous manipulation and data-driven AI research. Comparable as the leading well-funded industrial robotics AI research group with similar publication cadence.
• Boston Dynamics: Founded by the same person (Marc Raibert), Boston Dynamics is the commercial humanoid and quadruped robotics leader. It is highly comparable because it shares a founder, overlapping research DNA, and operates in the same intelligent-machines category — but it is now a Hyundai-owned commercial product company rather than a research-only institute.
• Toyota Research Institute: A large-scale, well-funded private research institute investigating robotics, ML, and energy. Directly comparable to RAI as a similarly-structured independent research organization with substantial budgets, robotics focus, and an emphasis on manipulation and learning-based control.
• Meta AI (Robotics / FAIR): Meta's robotics research arm (including projects like Habitat, ALOHA, and foundation-model efforts for manipulation) is highly comparable to RAI's dexterous manipulation and VLA-model research, though Meta operates at vastly larger scale and embeds the work inside a broader social-media company.

Others

• NVIDIA Robotics (Isaac Lab / GEAR): NVIDIA's robotics research and Isaac Lab simulation platform is referenced by RAI as training infrastructure and addresses the simulation-to-reality gap. Comparable as an adjacent enabling technology provider and research collaborator rather than a direct competitor.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks5 records

Key highlights7 records

Customer concentration

• Xx.com/rai_inst
• YouTubeyoutube.com/@rai-inst
• LinkedInlinkedin.com/company/rai-inst
• Instagraminstagram.com/robotics_and_ai_institute

Revenue estimate

Valuation estimate

Number of profiles

Profiles2 records

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about The RAI Institute (formally the Boston Dynamics Ai Institute)