On Wednesday, Boston Dynamics revealed a collaboration aimed at enhancing reinforcement learning capabilities within its electric Atlas humanoid robot. This alliance is formed with the Robotics & AI Institute (formerly known as The Boston Dynamics AI Institute).
Both entities were established by Marc Raibert, who was a MIT professor and Boston Dynamics’ CEO for three decades. The Institute, which came into existence in 2022, enables Raibert to pursue research that lays the groundwork for Boston Dynamics’ innovations.
Both organizations are associated with Hyundai, which acquired Boston Dynamics in 2021. Hyundai also supports the Institute financially, allowing Raibert the freedom to delve into more experimental, cutting-edge technologies, which may be difficult under a commercial umbrella. This initiative is akin to Toyota’s establishment of the Toyota Research Institute (TRI), which also collaborated with Boston Dynamics in October, centering on large behavior models (LBMs).
These partnerships aim to refine Boston Dynamics’ Atlas humanoid robot’s ability to learn new tasks. The specific focus with the Robotics & AI Institute centers on reinforcement learning, a technique that mimics the trial and error method prevalent in human and animal learning. Traditionally, reinforcement learning is a labor-intensive process; however, effective simulations have enabled multiple processes to occur simultaneously in virtual environments.
The collaboration between Boston Dynamics and the RAI Institute commenced earlier this month in Massachusetts. This is part of an ongoing series of partnerships that includes a joint initiative to create a reinforcement learning research kit for Boston Dynamics’ quadrupedal Spot robot, often referred to as the company’s robotic “dog.” The current research emphasizes transferring simulation-based learning into real-world applications and enhancing the Atlas robot’s movement and interaction within physical settings.
Regarding the latter, Boston Dynamics highlights capabilities such as “dynamic running and full-body manipulation of heavy objects.” These tasks necessitate the coordination of limbs, with the humanoid’s bipedal structure posing unique challenges and opportunities compared to Spot. Each action must also contend with varying forces, including balance, motion, force, and resistance.
Looking ahead, Raibert stated, “Our mission at RAI is to develop technology that empowers future generations of intelligent machines. Collaborating on Atlas with Boston Dynamics allows us to push the boundaries of reinforcement learning on arguably the most advanced humanoid robot available. This initiative is crucial for enhancing humanoid capabilities by not only broadening its skillset but also streamlining the acquisition of new skills.”
The announcement of this partnership comes just one day after Figure AI’s founder and CEO, Brett Adcock, revealed the company’s decision to end its collaboration with OpenAI in favor of establishing its own in-house AI models.
“We recognized that to effectively implement embodied AI at scale in the real world, vertical integration of robot AI is essential,” the executive told TechCrunch. “Outsourcing AI is not feasible for the same reason outsourcing hardware isn’t viable.”
Figure AI determined that the most suitable AI models for its humanoid robots are those developed internally. OpenAI’s strategy for embodied intelligence—referring to AI in physical forms—has been understandably less concentrated due to the company’s vast scope and size. This information also emerged alongside rumors suggesting that OpenAI is considering the development of its own humanoid robots.
Most companies venturing into the humanoid sector are creating their customized AI models. This is definitely true for Boston Dynamics, which boasts decades of experience in developing software tailored to its unique robotic systems. While the RAI Institute functions as a separate entity, both share the same founder, parent organization, and likely mutual objectives.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
