Boston Dynamics announced on Wednesday a new alliance with the Robotics & AI Institute (formerly known as The Boston Dynamics AI Institute). This collaboration aims to enhance the reinforcement learning capabilities of Atlas, Boston Dynamics’ advanced humanoid robot. The founder of both organizations is Marc Raibert, the ex-MIT professor who led Boston Dynamics as CEO for three decades.
The Robotics & AI Institute, established in 2022, provides Raibert a platform to further his research, which has been a cornerstone of Boston Dynamics’ advancements. Interestingly, both entities share connections with the South Korean automotive giant, Hyundai. Hyundai took over Boston Dynamics in 2021 and now also funds the Institute, which allows Raibert to delve into more exploratory and cutting-edge technologies than a conventional commercial company would typically allow.
This isn’t the first time Boston Dynamics has partnered with an external institute. In October, they announced a collaboration with Toyota Research Institute, focused on the use of large behavior models. The ultimate goal of these partnerships is to enhance the learning capabilities of Boston Dynamics’ Atlas humanoid.
The alliance with the Robotics & AI Institute is specifically centered on reinforcement learning, a learning method that works on a trial and error basis, much like how humans and animals learn. This technique is traditionally time-consuming, but the advent of effective simulation has facilitated multiple processes to occur simultaneously in a virtual environment.
The partnership between Boston Dynamics and the RAI Institute was officially launched earlier this month in Massachusetts. This alliance is the latest in a series of collaborations aimed at developing a reinforcement learning research kit for Boston Dynamics’ quadruped robot, Spot. The current focus is on transferring simulation-based learning to real-world scenarios and enhancing how Atlas navigates and interacts with physical environments.
Boston Dynamics aims to achieve “dynamic running and full-body manipulation of heavy objects” with Atlas. These tasks highlight the need for synchronization between the robot’s legs and arms. The bipedal form factor of the humanoid robot presents numerous challenges, but also offers several opportunities when compared to Spot. Every task Atlas undertakes is subjected to a range of forces, such as balance, force, resistance, and motion.
Raibert stated, “Our objective at RAI is to develop technology that empowers future generations of intelligent machines. Collaborating with Boston Dynamics on Atlas allows us to make progress in reinforcement learning on one of the most advanced humanoid robots available. This work will play a vital role in enhancing the capabilities of humanoids by not only increasing their skillset but also streamlining the process to acquire new skills.”
The announcement of this partnership comes a day after Brett Adcock, CEO of Figure AI, announced that his company is terminating its partnership with OpenAI in favor of developing its own in-house models. Adcock explained that to address embodied AI at scale in the real world, it’s necessary to integrate robot AI vertically. He added that the best AI models for their humanoid are those specifically developed for its robots in-house.
Most companies in the humanoid space, including Boston Dynamics, are developing their own unique AI models. With decades of experience in creating software for its unique robotic systems, Boston Dynamics continues to lead in innovation. Despite the RAI Institute being a separate entity, both share common goals under the same parent company and founder.
