The Massachusetts Institute of Technology has come up what it thinks is a better way to teach robots general purpose skills. Derived from LLM techniques, the method provides robot intelligence access to an enormous amount of data at once, rather than exposing it to individual programs for specific tasks. Faster and more cost efficient, the approach has been referred to as a “brute force” approach to problem-solving, and machine learners have taken to it in lieu of individualized, task-specific “imitation learning.” Early tests show it outperforming traditional training by more than 20 percent under simulation and real-world conditions.

“In robotics, people often claim that we don’t have enough training data. But in my view, another big problem is that the data come from so many different domains, modalities, and robot hardware,” Lirui Wang, an MIT electrical engineering and computer science graduate (EECS) student told MIT News, adding that “our work shows how you’d be able to train a robot with all of them put together.

Wang is lead author of a scientific paper on the research, conducted with Carnegie Mellon University and partial funding from Toyota. “Wang’s co-authors include fellow EECS graduate student Jialiang Zhao; Xinlei Chen, a research scientist at Meta; and senior author Kaiming He, an associate professor in EECS and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL),” reports TechCrunch.

“Our dream is to have a universal robot brain that you could download and use for your robot without any training at all,” CMU Associate Professor David Held told TechCrunch, explaining the hope is that “scaling leads to a breakthrough in robotic policies, like it did with large language models.”

The model, Heterogeneous Pre-trained Transformers (HPT), allow robots to absorb and adapt to executing a wide range of tasks. Lack of diverse learning has to date been the biggest robot challenge, with new situations or unexpected obstacles stupefying mechanical assistants.

“The research could lead to a future where robots are not just specialized tools but flexible assistants that can quickly learn new skills and adapt to changing circumstances, becoming truly general-purpose robotic assistants,” writes TechSpot.

“At the heart of the HPT architecture is a transformer, a type of neural network that processes inputs from various sensors, including vision and proprioception data, and creates a shared ‘language’ that the AI model can understand and learn from,” TechSpot explains.