Researchers at the University of California, Berkeley, have made significant strides in the field of humanoid robotics, creating a versatile control system for these robots to navigate various terrains and obstacles adeptly. Drawing inspiration from the deep learning frameworks that revolutionized large language models (LLM), this AI system hinges on a core principle: utilizing recent observations to predict future states and actions.
This groundbreaking system was trained entirely in simulation but demonstrates robust performance in unpredictable real-world settings. Analyzing its past interactions, the AI dynamically refines its behavior to effectively tackle novel scenarios it never encountered during its training phase. This advancement in AI technology propels responsible AI forward, promising a future where humanoid robots become valuable assistants, well-equipped to navigate the world and assist in various physical and cognitive tasks.
Traditional control systems in robotics have been notoriously inflexible, often designed for specific tasks and unable to cope with the unpredictability of real-world terrains and visual conditions. This rigidity limits their utility, confining them to controlled environments. The new control system created by the scientists at U.C. Berkeley, however, challenges this limitation.
Deployed on Digit, a full-sized, general-purpose humanoid robot, the system demonstrates remarkable outdoor walking capabilities. It navigates reliably across everyday human environments such as walkways, sidewalks, running tracks and open fields, and handles various terrains, including concrete, rubber, and grass, without falling.
At the heart of this system is a “causal transformer,” a deep learning model that processes the history of proprioceptive observations and actions. The causal transformer excels at discerning the relevance of specific information, such as gait patterns and contact states, to the robot’s observations. This advancement in transformer technology enables the robot to predict the consequences of actions with high precision and modify its behavior to attain more favorable future states.
This development in AI is a leap forward in robotic control systems. The researchers at U.C. Berkeley have created an adaptable system that allows humanoid robots to navigate various terrains with ease. With the continued integration of transformers in robotics, we are moving closer to a future where humanoid robots can have a significant impact on our daily lives.