Boston Dynamics announced a new partnership on Wednesday aimed at enhancing the reinforcement learning capabilities of its electric Atlas humanoid robot. This collaboration is established with the Robotics and Artificial Intelligence Institute (RAI Institute), formerly known as the Boston Dynamics Artificial Intelligence Institute. The institute was founded in 2022 by former MIT professor and Boston Dynamics' former CEO Marc Raibert, with the goal of continuing the foundational research that supports Boston Dynamics.
The partnership between Boston Dynamics and the RAI Institute is supported by Hyundai Motor Company. Hyundai acquired Boston Dynamics in 2021 and provided funding for the institute, allowing Raibert to explore more experimental and cutting-edge technologies that are difficult to pursue in commercial companies. The model of the RAI Institute is similar to that of the Toyota Research Institute (TRI), which also partnered with Boston Dynamics in October to focus on using large behavioral models (LBMs).
These two collaborations aim to enhance the ability of Boston Dynamics' electric Atlas humanoid robot to learn new tasks. The collaboration with the RAI Institute focuses on reinforcement learning, a method of learning through trial and error, similar to the learning processes of humans and animals. Although traditional reinforcement learning usually requires a significant amount of time, the emergence of effective simulation technologies allows many processes to occur simultaneously in a virtual environment.
The collaboration between Boston Dynamics and the RAI Institute launched earlier this month in Massachusetts. This partnership is the latest outcome of their many collaborations, having previously developed a reinforcement learning research toolkit for Boston Dynamics' quadruped robot, Spot. The new collaboration primarily focuses on how to transfer simulated learning to real-world environments and improve Atlas's mobility and interaction capabilities in human physical environments.
In terms of the latter, Boston Dynamics mentioned "dynamic running and full-body manipulation of heavy objects," both of which require synchronization between limbs. Compared to Spot, the bipedal form of the humanoid robot faces unique challenges and opportunities in movement, with each activity influenced by various forces such as gravity, strength, resistance, and motion.
On a broader level, Raibert noted in a statement: "Our goal at RAI is to develop technologies that will enable future intelligent machines. Collaborating with Boston Dynamics on Atlas allows us to advance reinforcement learning progress on what is undoubtedly the most complex humanoid robot. This work will be crucial in enhancing the capabilities of humanoid robots, not only by expanding their skill sets but also by streamlining the process of acquiring new skills."
The announcement of this collaboration comes as Brett Adcock, founder and CEO of Figure AI, revealed that the company is abandoning its partnership with OpenAI to develop its own models. He stated that addressing large-scale embodied AI problems in the real world requires vertically integrated robotic AI, rather than relying on outsourcing.
Key Points:
🤖 Boston Dynamics partners with RAI Institute to enhance Atlas humanoid robot's reinforcement learning capabilities.
🔍 Reinforcement learning is conducted through trial and error, similar to the learning processes of humans and animals.
🚀 The goal is to empower future intelligent machines by collaboratively developing new technologies and skills.
