Boston Dynamics and Toyota Research Institute (TRI) are joining forces to supercharge the advancement of humanoid robots. Their goal is to create smarter robots capable of solving real-world problems quickly and efficiently.
With this partnership, TRI will gain access to Boston Dynamics’ cutting-edge humanoid robot, Atlas. TRI will test its advanced AI models on this highly capable platform, focusing on enhancing the robot’s ability to handle multiple tasks and perform complex manipulations.
The data collected from these tests will help TRI further refine its AI models, allowing for faster problem-solving and more efficient robot performance.
In return, Boston Dynamics will benefit from Toyota’s expertise in AI development. Toyota’s Large Behavior Models (LBMs) will improve Atlas’ dexterity and decision-making, enabling the robot to perform a wider range of tasks.
Boston Dynamics, which is owned by Hyundai, plans to use Toyota’s AI to explore new innovations in robotics and ensure that Atlas can adapt to various real-world environments.
“There has never been a more exciting time for the robotics industry, and we look forward to working with TRI to accelerate the development of general-purpose humanoids,” Boston Dynamics CEO Robert Playter said in a press release. “This partnership is an example of two companies with a strong research-and-development foundation coming together to work on many complex challenges and build useful robots that solve real-world problems.”
Gill Pratt, chief scientist for Toyota and CEO of TRI, added: “Recent advances in AI and machine learning hold tremendous potential for advancing physical intelligence. The opportunity to implement TRI’s state-of-the-art AI technology on Boston Dynamics’ hardware is game-changing for each of our organizations as we work to amplify people and improve quality of life.”
Atlas has been in development since 2013, when it was introduced as part of the DARPA Robotics Challenge seeking innovations in disaster response. The robot was designed to complete tasks like turning valves and operating machinery in dangerous environments. It was equipped with advanced sensory systems, including LIDAR and stereo vision, to perceive and interact with its surroundings effectively.
With the robot’s first major upgrade in 2015, Atlas became more agile and could traverse rough terrain without being tethered to a power source. At first, Atlas required a human operator, but it gained increasing levels of autonomy through subsequent revisions. It evolved to perform tasks independently, making real-time decisions based on environmental conditions.
By 2016, a newer and bedazzled version of Atlas went viral performing parkour-like movements. Millions of people watched as the robot jumped obstacles, balanced on one leg, and quickly recovered from falls. It also inspired parodies.
Subsequent upgrades improved the robot’s abilities to handle real-world tasks like lifting boxes and manufacturing processes.
Boston Dynamics retired the hydraulic humanoid robot in April 2024 after 11 years of achievements. The company has since unveiled a new, fully electric version of Atlas designed for real-world applications.
The new Atlas features electric actuators instead of hydraulics, improving efficiency, control, and reducing complexity. Electric systems are quieter and more energy-efficient, essential for robots performing continuous tasks.
Boston Dynamics has also enhanced its software, incorporating advanced AI, machine learning, reinforcement learning, and computer vision to help Atlas understand and adapt to complex situations, boosting its real-world utility.
The company emphasized in a blog post that commercializing advanced robots like Atlas requires patience.
The Toyota Research Institute, founded in 2016, focuses on artificial intelligence, robotics, and materials science. The institute says robots should amplify human abilities rather than replace them.
Around 2017, the institute shifted more attention toward robotics for home and personal assistance.
Last year, the institute introduced a new way for robots to learn complex tasks quickly through demonstration. With the system, robots can learn new skills like peeling veggies or flipping pancakes in a single afternoon. Human operators show robots how to complete tasks, and the robots learn from the demonstrations.
An AI technique called Diffusion Policy helps the robots understand and complete the tasks by planning actions step by step. This approach allows robots to handle tasks in multiple ways, like picking up an object from different angles.
By inheriting these insights, Atlas can better perform tasks autonomously, adapt to new environments, and handle versatile challenges.
Earlier this year, the research hub introduced its Punyo humanoid robot. Punyo has soft tactile sensors covering its hands, arms and chest to allow for safe, flexible and reliable manipulation. Punyo’s design uses bubble-like grippers with tactile sensors to evenly distribute forces and prevent slipping.
The robot handles large-scale tasks by using its whole body, not just its hands. The institute tests these features using teleoperation and simulations to teach Punyo complex tasks.
Boston Dynamics is accelerating development of Atlas as a growing array of players race to mass produce their humanoid robots for industrial and domestic applications.
Just last week, Tesla’s Optimus humanoid robots stole the show as the EV automaker unveiled its much-anticipated Robotaxi.
Plans call for limited production of Tesla humanoids in 2025 and wider scale production in 2026. Musk expects the robot to reach a production scale of 1 million units per year within five to six years.
California-based Figure is another major contender. In January, several of the OpenAI-backed startup’s Figure 02 robots will start working full-time at the BMW plant in Spartanburg South Carolina after a successful trial earlier this year.
Like Tesla, Figure plans to scale production with the third iteration of its robot starting in 2025, with the goal of eventual mass production and deployment. Figure founder Brett Adcock recently said the Figure 3 is designed for high-rate manufacturing and large-scale use in industrial settings. Similar to Tesla, Figure plans to market its robots for domestic use after they prove themselves in factories.
Another OpenAI-backed startup, 1X Technologies, intends to introduce its NEO humanoid robots into homes instead of starting in industry. When the much-hyped robot enters homes, it will likely rely on human operators to start. The company plans to use human teleoperation as a stand-in for autonomy at first. So the NEO can still perform chores in a household but it’ll be fulfilled by a human operator who’s remotely controlling it.
Agility Robotics plans to start manufacturing its humanoid robot, Digit, at its new RoboFab facility in Salem, OR.
GXO Logistics, one of the world’s largest logistics and supply chain providers, has already put the bipedal Digit to work as part of a multi-year agreement. The robot is primarily used for logistics tasks today but the company plans to expand its capabilities as technology advances.
There’s also an onslaught of advanced AI robots emerging from China like the Fourier GR-2, the XPeng PX5, the Xiaomi CyberOne, and Unitree’s H1 and G1. The short king G1 is especially disruptive with its $16K price point.