Boston Dynamics announced a new partnership on Wednesday with the Robotics & AI Institute (RAI Institute) to enhance the reinforcement learning capabilities of its electric Atlas humanoid robot. This collaboration aims to improve how Atlas learns new tasks through more efficient and advanced methods.
The RAI Institute, previously known as The Boston Dynamics AI Institute, was founded by Marc Raibert, a former MIT professor and the former CEO of Boston Dynamics for 30 years. Raibert established the Institute in 2022 to continue the cutting-edge research that formed the foundation of Boston Dynamics’ innovative robotics.
Both organizations have strong ties to Hyundai, which acquired Boston Dynamics in 2021. Hyundai also funds the RAI Institute, giving Raibert the freedom to pursue more experimental and pioneering technologies beyond the scope of a commercial company. The RAI Institute’s mission is similar to that of Toyota’s Research Institute (TRI), which is also focused on pushing the boundaries of robotics and AI. In fact, Boston Dynamics and TRI formed their own partnership last October, centered around the use of large behavior models (LBMs).
The Boston Dynamics/RAI Institute partnership aims to enhance the learning processes of the Atlas humanoid robot, specifically focusing on reinforcement learning. This learning technique is based on trial and error, mirroring how humans and animals acquire new skills. While reinforcement learning has traditionally been slow and labor-intensive, advancements in simulation technologies have enabled more efficient virtual training, allowing robots to learn multiple tasks simultaneously in a digital environment.
The partnership officially kicked off earlier this month in Massachusetts and marks the continuation of several joint efforts between the two organizations. One such effort includes the development of a reinforcement learning research kit for Boston Dynamics’ quadrupedal robot, Spot. The new initiative will focus on transferring simulation-based learning to real-world environments, while also improving how Atlas moves and interacts physically.
Boston Dynamics points to specific challenges the humanoid robot faces, such as dynamic running and full-body manipulation of heavy objects—tasks that require precise coordination between the legs and arms. The bipedal design of Atlas presents unique challenges compared to Spot, with every movement being influenced by factors like balance, force, resistance, and motion.
Marc Raibert shared his vision for the collaboration, saying, “Our aim at RAI is to develop technology that enables future generations of intelligent machines. Working on Atlas with Boston Dynamics allows us to make breakthroughs in reinforcement learning with one of the most advanced humanoid robots available. This research will be crucial for advancing humanoid capabilities, expanding their skill sets, and streamlining the process of teaching robots new abilities.”
The announcement of this partnership comes just a day after Brett Adcock, the CEO of Figure AI, revealed that his company is ending its collaboration with OpenAI in favor of developing proprietary AI models for its humanoid robots. Adcock explained that to solve the challenges of embodied AI at scale, it’s necessary to vertically integrate AI and robotics, rather than relying on external solutions like OpenAI’s models.
This trend of developing bespoke AI systems for humanoid robots is common across the industry. Like Figure AI, Boston Dynamics has extensive experience in creating AI software tailored specifically for its unique robotic systems. Although the RAI Institute is technically a separate entity, both organizations share a common founder, a parent company, and a shared mission to advance the field of robotics and AI.
By Impact Lab
