Boston Dynamics and Toyota Research Institute (TRI) have announced a groundbreaking collaboration to infuse AI-based robotic intelligence into the Atlas humanoid robot. The partnership will utilize TRI’s expertise in large behavior models (LBMs), similar to the large language models (LLMs) used in platforms like ChatGPT.
TechCrunch recently visited TRI’s Bay Area campus to explore its advances in robot learning. At last year’s Disrupt conference, TRI’s head, Gill Pratt, demonstrated how the institute achieved 90% accuracy in teaching robots to perform tasks like flipping pancakes through overnight training.
Pratt explained the significant shift in machine learning. "Previously, machine learning required millions of training cases, which is impractical for physical tasks due to wear and tear on machines. Now, we need just dozens of diverse cases for effective training," Pratt noted. This breakthrough has drastically reduced the amount of training necessary for robots to perform complex physical tasks.
Boston Dynamics is a perfect match for TRI on the hardware side. While the company has made strides in robot software and AI, the challenge of teaching robots to perform fully autonomous, intricate tasks is monumental. The Spot robot, developed by Boston Dynamics, is a prime example of the company's success in utilizing AI for automation.
“There has never been a more exciting time for the robotics industry," said Boston Dynamics CEO Robert Playter. "Our partnership with TRI will accelerate the development of general-purpose humanoid robots. This collaboration brings together two companies with strong R&D foundations to tackle complex challenges and build useful robots that solve real-world problems.”
In April, Boston Dynamics revealed the design for its electric Atlas humanoid, marking a shift from its older hydraulic model. Though the robot has remained mostly out of the spotlight, a video surfaced in August showcasing Atlas doing pushups, demonstrating its impressive strength.
Boston Dynamics faces competition from companies like Agility, Figure, and Tesla, which are also developing humanoid robots. However, while competitors are building their AI teams in-house, the Boston Dynamics-TRI partnership stands out, especially considering both organizations are backed by automotive giants Hyundai and Toyota — direct competitors in the automotive sector.
Boston Dynamics' AI Institute (formerly the Boston Dynamics AI Institute), led by founder and former CEO Marc Raibert, operates independently from its parent company. While the institute is still young and in the process of building out its team, it is expected to play a crucial role in advancing robotic AI research. TRI, on the other hand, has shifted its focus away from hardware to concentrate more on the AI side of robotics.
The ultimate goal of these efforts is to develop a general-purpose machine — a system that can learn and perform any task a human can, and possibly even surpass human capabilities. While robot hardware has advanced, achieving general intelligence remains one of the most challenging problems in artificial intelligence.
The development of SDKs (software development kits) has significantly increased the range of tasks robots like Boston Dynamics’ Spot can perform. However, true artificial general intelligence (AGI), which would allow robots to handle any task autonomously, is still a long way off — if it's even possible.
As Boston Dynamics and TRI continue their work, the collaboration will push the boundaries of what's possible in robotics and AI, bringing us closer to a future where humanoid robots with general intelligence are part of everyday life. The robotics world is witnessing an exciting era of innovation, with companies like Boston Dynamics and TRI leading the charge.
Post a Comment