The humanoid robot Figure can now learn to perform any manual task simply by watching a human worker at work. This may well mean the disappearance of many jobs… but it’s also very good news!
A “ ChatGPT moment “For humanoid robotics? That’s what Figure CEO Brett Adcock promised over the weekend of January 6, 2024.
A way of announcing a worthy innovation of the OpenAI chatbot that sent a seismic shock through the world last year… so, what’s it all about?
We finally have the answer. The robot manufactured by the Californian company is now capable of observing humans performing tasksto understand how to do them, and then immediately begin to perform them fully autonomously.
Learning by looking: an essential capability for humanoid robots
This is a major milestone has been reached in the field of humanoid robots, as these machines, designed to be generalists, will have to be capable of doing all kinds of jobs.
In order to be useful, they must be able to understand all the toolsdevices, objects, techniques and objectives that humans use to achieve their ends. In addition, they will need to be flexible and adaptable to our huge variety of working environments.
Clearly, if it is necessary for a team of programmers to learn robots to perform each job one by one, this technology will be totally useless.
It is essential that humanoids can learn by observing. They will need multimodal AI capable of watching and interpreting videos, then controlling a robotic body to replicate what they see.
For several years now researchers have been trying to create such an AIsuch as the Google Deepmind researchers who recently unveiled a system capable of learning by imitating.
Similarly, in September 2023, Toyota unveiled its “large behavior model: the equivalent of a large language model (LLM) such as GPTbut for action.
However, the Japanese automotive giant had merely applied this innovation to robotic arms in a research center. However, companies such as FigureTesla and its Optimus, or Agility and its Apptronik, much sought-after by NASA, have much broader ambitions.
Figure opens the Pandora’s box of robotics
As you can read in our feature on the humanoid robotics revolution, these companies and many others are all aiming to create fully autonomous bipedal machines capable of performing any manual task in place of humans.
In addition, it not just research projects. Their aim is to start marketing such robots in the coming months. By the end of 2023, after only 12 months of development, Figure had already taught its humanoid to walk. and began deploying it internally.
That’s why this new announcement from the company represents a pivotal moment for the entire industryand, by extension, for humanity.
Of course, the demonstration itself is unlikely to knock you off your chair. It simply shows the Figure robot using a Keurig coffee machinewith a cup already inside.
In response to a simple oral order, he opens the hood, places a coffee capsule insideHe then closes the machine and presses the button. He then lets the person who asked him to take the cup himself.
The coffee machine and the human are still essential to the process, but that’s not the point. What’s important is that this robot needed just 10 hours to analyze a video and learn to perform the task on its own.
Figure 01 robots can share their learning with each other
It has added this action to its internal librarylibrary, allowing it to be transfer it to any other robot Figure to share this learning with it via “swarm learning”…
Do you understand what this means? From now on, every time a Figure robot learns a new task, all the others Figure robots will learn it simultaneously.
In a few years, a few months, these machines will be able to learn to perform more actions and manual tasks than the majority of human beings.including the most resourceful among us.
Coming soon, the robot will be able to retrieve the cup itself in the kitchen, check that the machine is plugged in and sufficiently full of water, prepare the coffee and bring it to your desk (without spilling it on you and scalding you, thanks to the Constitution Robot that Google has just created).
This complex task requires both action capabilities and AI of the Large Language Model type to break it down into simple steps. And this is just the beginning.
This is indeed a crucial step forward, and things are going to move very fast from now on. Other companies will soon be developing similar technologies, and robots will soon be ubiquitous in our businesses and in everyday life…
This innovation will totally transform our planet, and make ChatGPT look like a fun gadget. At the end of the road, is quite simply the end of human work. Are you ready for it?