Recently, Boston Dynamics and the Toyota Research Institute (TRI) announced a collaborative plan to integrate AI technology into the electric humanoid robot Atlas. This partnership will leverage TRI's research in Large Behavior Models (LBMs), which operate on principles similar to the Large Language Models (LLMs) familiar to us, such as ChatGPT.
Last September, we visited TRI's campus in the Bay Area and learned about their achievements in robot learning. TRI's head, Gill Pratt, revealed at last year's Disrupt conference that their lab had achieved a 90% accuracy rate in executing household tasks, such as flipping pancakes, through overnight training.
Pratt stated that traditional machine learning requires millions of training samples, but such training time is infeasible for physical tasks, as the machine may be damaged before reaching ten thousand repetitions. The current trend is to train with just dozens of samples, leveraging their diversity, sometimes even fewer.
In terms of hardware, the collaboration between Boston Dynamics and TRI complements each other well. Boston Dynamics has also made progress in software and AI, but teaching robots to autonomously perform complex tasks remains a significant challenge.
Robert Playter, CEO of Boston Dynamics, stated in the announcement: "This is the most exciting time in the robotics industry, and we look forward to collaborating with TRI to accelerate the development of general-purpose humanoid robots." This collaboration showcases two companies with strong R&D foundations joining forces to tackle complex challenges, aiming to create practical robots that can solve real-world problems.
In April of this year, Boston Dynamics first showcased the design of the electric Atlas, marking the humanoid robot's liberation from the constraints of previous large hydraulic versions. Although we have seen few related videos since then, in August, TechCrunch released a short video showing Atlas's ability to do push-ups, demonstrating its remarkable strength.
Boston Dynamics faces significant competition in developing humanoid robots, with companies like Agility, Figure, and Tesla opting to build their own AI teams. This collaboration between Boston Dynamics and TRI is particularly noteworthy as both companies are operated by Hyundai and Toyota, direct competitors in the automotive field.
Additionally, Boston Dynamics has its own research branch, the AI Institute (formerly the Boston Dynamics AI Institute), which, although managed by Boston Dynamics founder Marc Raibert, remains independent from Boston Dynamics. Meanwhile, TRI has gradually reduced its investment in hardware.
The goal of this collaboration is to create a truly general-purpose machine, one that can learn and complete all tasks as humans do, and even more. Although we have made significant progress in robot hardware, achieving near-true general intelligence remains a formidable challenge.
Key Points:
🤖 Boston Dynamics and Toyota Research Institute collaborate to introduce AI technology into the Atlas humanoid robot.
🌟 TRI has made significant progress in robot learning, achieving a 90% accuracy rate in household tasks.
🚀 The goal of the collaboration is to develop a general-purpose robot capable of learning and performing tasks like a human.