AI advancements are speeding up the training of robots and helping them to do new tasks almost immediately.
WHO?
Amazon, Covariant and Robust are all examples of companies that have developed robots.
When?
And now
Generative artificial intelligence is changing the way robots are taught. It’s now obvious how we can finally build the kind of capable robots that were once the stuff of science-fiction.
Researchers in robotics are not strangers to artificial intelligent–it’s been helping robots detect objects on their path for years. Roboticists were amazed by the progress made in large language modeling a few decades ago. The makers of these models could feed them massive quantities of text — books, poems, manuals — and then fine-tune the models to generate text on prompts. Explore the complete list of 10 Breakthrough technologies for 2025.
The idea to do the same with robotics was tantalizing, but incredibly complex. It’s one to use AI to create sentences for a screen. But it’s another to use it to teach a robot how to move and do useful things. Now, roboticists are making major advances in this pursuit. One was to figure out how to combine data from different sources and make it useful and legible for a robot. As an example, let’s say you were to wash dishes. You can collect data by having someone wash dishes while wearing sensors. You can then combine this data with teleoperational data from a person performing the same task using robotic arms. You can also scrape images and videos from the internet of people doing dishes.
By combining these data sources into a new AI-model, it is possible to train robots that, while not perfect, have a huge head start over those who are trained using more manual methods. AI models can improvise and predict what a robot should do in the real world easier when they see so many different ways a task can be performed.
This breakthrough is set to redefine the way robots learn. These advanced training methods are already used by robots in commercial spaces such as warehouses. The lessons learned from these experiments could be the foundation for smart robots at home.