On Wednesday, Boston Dynamics announced a partnership designed to bring improved reinforcement learning to electric atlas humanoid robots. The partnership is at the Robotics & AI Institute (RAI Institute), formerly known as the Boston Dynamics AI Institute.
Both organizations were founded by Marc Raibert, a former MIT professor who served as CEO of Boston Dynamics for 30 years. Founded in 2022, the institute will allow Raibert to continue his research that serves as the basis for Boston Dynamics.
Both have something to do with Hyundai. The Korean automaker acquired Boston Dynamics in 2021. Hyundai also funds the Institute, and Raybert gives free reins to explore more experimental and bleeding edge edge technologies than commercial companies can. The institute reflects Toyota's creation for the Toyota Institute, and announced its own partnership with Boston Dynamics in October, focusing on the use of large-scale behavioral models. (LBMS).
The Twin Partnership is designed to improve the way Boston Dynamics' Electric Atlas Humanoids learn new tasks. The Robotics & AI Institute's deal focuses specifically on reinforcement learning, a way that works through trial and error, as well as how both humans and animals learn. Although reinforcement learning has traditionally been very time-consuming, the creation of effective simulations allowed many processes to be performed at once in a virtual setting.
The Boston Dynamics/RAI Institute Union began earlier this month in Massachusetts. This is the latest in many collaborations between the pair, including a collaborative effort to develop a Quadrupedal Spot Robot reinforcement learning research kit by Boston Dynamics (which is the familiar robot “dog”). The new work focuses on transferring simulation-based learning to a real-world setting and improving how the company's humanoid rasses move and interact with the physical environment.
Boston's dynamics associated with the latter point to “dynamic running and full-body manipulation of heavy objects.” Both are examples of actions that require leg and arm synchronization. Humanoid bipedal form factors present many unique challenges and opportunities when compared to spots. All activities are subject to a wide range of forces, including balance, force, resistance, and movement.
Raibert said, “The purpose of RAI is to develop technology that enables future generations of intelligent machines. By working on Atlas using Boston Dynamics, we can now use accelerating learning with perhaps the most sophisticated humanoid robots. It can advance. This work will not only expand your skill set, but also streamline the process to achieve new skills, but also play a key role in advancing humanoid capabilities.”
News of the partnership arrived the day after AI founder and CEO Brett Adcock announced that the Bay Area Company was abandoning its partnership with Openai in support of developing its own in-house model.
“We've discovered that in order to solve real-world AI at scale, we need to vertically integrate robotic AI,” the executive told TechCrunch. “You can't outsource AI for the same reasons as you can't outsource your hardware.”
The figure determined that the best AI model for the humanoid was developed specifically for the robot. Openai's approach to embodied intelligence – the meaning of physical forms – is of course less focused given the size and range of ChatGpt makers. The news also reached the heels that Openry is exploring the creation of its own humanoid robots.
Most companies involved in the humanoid space are working on their own custom AI models. It certainly applies to Boston Dynamics. Boston Dynamics has decades of experience developing software to run on its own robotic systems. The RAI Institute is a separate technical organization, but both share parent companies, founders and possibly common goals.
