Two of the world’s largest automakers, Hyundai and Toyota, are joining forces to conquer the future of robotics. In a landmark collaboration announced on October 16, 2024, Hyundai-owned Boston Dynamics and the Toyota Research Institute (TRI) agreed to combine their strengths. The partnership aims to merge the physical prowess of Boston Dynamics’ new electric Atlas robot with the advanced artificial intelligence developed by Toyota. By integrating top-tier hardware with “Large Behavior Models,” the companies hope to create general-purpose humanoid robots that can learn complex tasks and work alongside humans, directly challenging competitors like Tesla’s Optimus.
The Deal: Boston Dynamics Meets Toyota Research Institute
The agreement brings together two distinct leaders in the robotics field. Boston Dynamics, acquired by Hyundai Motor Group in 2021, is renowned for building the world’s most dynamic and agile robots. Their machines, such as Spot and Atlas, are famous for dancing, backflipping, and navigating rough terrain with uncanny balance.
On the other side is the Toyota Research Institute (TRI), the advanced research arm of Toyota Motor Corporation. TRI has spent years focusing on the “brain” of the robot – specifically, how to teach machines to understand and manipulate their environment using artificial intelligence.
Under the new partnership, TRI will gain access to the new fully electric Atlas robot. In return, Boston Dynamics will utilize TRI’s Large Behavior Models (LBMs) to give their robots more autonomy. The project will be co-led by Scott Kuindersma of Boston Dynamics and Russ Tedrake of TRI, with research teams working together in Boston. Their immediate goal is to use the Atlas platform to collect data and train AI models to handle fundamental humanoid skills, such as bimanual manipulation (using two hands together) and whole-body control.
Why This Matters: A “Game-Changing” Collaboration
For decades, robots have been stuck in factories, performing repetitive motions that must be strictly programmed by code. Making them “general-purpose” – capable of doing laundry, stocking diverse shelves, or helping in a hospital – has been nearly impossible because the real world is too messy to hard-code.
This partnership addresses that bottleneck. Gill Pratt, Toyota’s chief scientist, called the collaboration “game-changing.” It signals a shift from manually programming robots to letting them learn.
Speed of Development
By sharing resources, both companies can move faster. Boston Dynamics doesn’t have to build a foundational AI model from scratch, and Toyota doesn’t have to build a new humanoid body. This allows them to bypass years of independent R&D.
Scaling Data
AI needs massive amounts of data to learn. The electric Atlas provides a reliable, highly capable physical platform to run experiments and generate the real-world data needed to train smart AI models. This data-driven approach is similar to how Physical Intelligence is training its own generalist robots.
Rising Competition
The race is heating up. Tesla plans to deploy its Optimus robots in factories by 2025. Other startups like Figure and Apptronik are also moving quickly. This alliance creates a formidable competitor that combines the manufacturing scale of Hyundai and Toyota with deep technical expertise.
The Technology: Hardware Meets “Large Behavior Models”
The Body: Electric Atlas
In April 2024, Boston Dynamics retired its famous hydraulic Atlas and unveiled a fully electric successor. This new machine is stronger, quieter, and more reliable than the old version. It features swiveling joints that allow it to turn its limbs in ways humans cannot, maximizing efficiency. It is designed for real commercial work, not just research demos.
The Brain: Large Behavior Models (LBMs)
Toyota Research Institute is a pioneer in Large Behavior Models. These are similar to the Large Language Models (LLMs) that power chatbots like ChatGPT, but instead of learning text, they learn physical actions.
- Diffusion Policy: TRI uses a technique called diffusion policy, a form of generative AI. It allows a robot to watch a human demonstrate a task – like whisking eggs or peeling vegetables – and learn the skill overnight.
- Generalization: The goal is for the robot to understand the concept of a task. If it learns to pick up a box, it should eventually be able to pick up a crate or a basket without being reprogrammed.
- Autonomy: By running TRI’s models on the Atlas robot, the machine can perceive its surroundings and make decisions in real-time, adjusting its grip or balance as needed without constant human guidance.
Sources & related information
Toyota Research Institute – Toyota Joins With Hyundai’s Boston Dynamics on AI-Powered Robots – 2024
Toyota announces the partnership to pair its Large Behavior Model expertise with the Atlas robot to accelerate general-purpose humanoid development.
KED Global – Hyundai-owned Boston Dynamics, Toyota team up for humanoid robots – 2024
Reports on the strategic alliance between the world’s No. 1 and No. 3 auto groups to challenge competitors like Tesla and Figure AI.
Boston Dynamics – Atlas – 2024
Details on the fully electric Atlas robot, designed for real-world applications with advanced strength, agility, and swiveling joints.
Toyota Research Institute – Large Behavior Models – 2024
Explains TRI’s work on LBMs, which use generative AI to teach robots new dexterous skills from human demonstration.
0 Comments