Will autonomous humanoid robots soon occupy our homes? – rts.ch

Will autonomous humanoid robots soon occupy our homes? – rts.ch
Will autonomous humanoid robots soon occupy our homes? – rts.ch

After developing systems like ChatGPT, capable of interacting with us by generating text, also through voice, scientists are now trying to equip humanoids with similar artificial intelligence capabilities to make them more autonomous in their movements.

“Hello how are you today?” Nadine sits at her table, her hands motionless and her gaze lost into space. Quite astonishingly, she calls out to the person and answers all the questions with a monotonous and choppy, but understandable voice.

Nadine is a social, humanoid robot. It bears a striking resemblance to the researcher who created it, Nadia Thalmann, director of MIRALab at the University of Geneva. “Nadine has been successfully tested in retirement homes, particularly in Singapore. She can chat using an artificial language model, such as ChatGPT,” explains Nadia Thalmann in the 7:30 p.m. of RTS.

Multiple biases

Unveiled at the end of 2022, this type of text production system can still be improved, since it sometimes generates meaningless content, which scientists call “hallucinations”. This “artificial intelligence” is also perverted by multiple biases linked to gender, political or social tendencies.

But for Janet Adams, chief operating officer of SingularityNET, which develops social robots, the future looks bright. “These developments will not happen gradually, but rather in the form of a breakthrough, when we have the technologies that will erase the current limitations of ChatGPT systems, so that robots become truly creative and learn from their environment. “

“We believe this paradigm shift will happen in the next 18 to 24 months. It will truly unlock the power of robots across the planet!”

Empower movements

However, roboticists are already working on the next step: equipping mobile humanoid robots, often bipedal, with artificial intelligence systems similar to those generating language, but with the objective of empowering these machines in their movements, particularly according to commands they receive from humans.

We could then ask our robot companion to open the fridge and literally cook a dish with the elements it finds there, without knowing them in advance.

In recent months, several companies, from Tesla to OpenAI via Boston Dynamics, have presented their latest humanoid robot, each time worthy of the best science fiction films, but nevertheless very real.

External content

This external content cannot be displayed because it may collect personal data. To view this content you must authorize the category Social networks.

Accept More information

In a video, we see a robot, named FigureO1, answering the question “Can I have something to eat?” posed by a human, by a movement of his robotic arm grabbing an apple. However, it is difficult to know to what extent the whole thing was pre-programmed.

External content

This external content cannot be displayed because it may collect personal data. To view this content you must authorize the category Social networks.

Accept More information

“I like this idea of ​​inserting ChatGPT-type systems into robots,” comments Yuichiro Kawasumi, chief engineer of the Japanese company Kawada Technologies Inc. “Our robots display very high performance, but they do not have enough of ‘brain’ Before, we had to program the robots, which were often used in factories. But it would be nice if they acquired more flexible functions, to interact with people.

Differentiate machine from human

A dizzying prospect for Azeem Azhar, an expert in the field of tomorrow’s technologies, founder of the blog Exponential View: “We humans appreciate being able to talk to these machines in a… human way. But if we insert such systems into robots, we have to be very careful to make sure everyone understands who is a human being and what is a machine.”

It has been around twenty years since we were able to show that machines are in reality capable of carrying out absolutely all the tasks of our daily lives as humans.

Davide Scaramuzza, director of the Robotics and Perception Group at the University of Zurich

Making robots fully autonomous, however, is far from trivial, as Nadia Thalmann explains. “Nadine is equipped with software. But when we talk about a humanoid robot, we talk about the physics of the body. But a robot is extremely simple: it has cables, motors and that’s all. It remains extremely far from the human body, from its highly evolved physiology. It is therefore very difficult to wait a long time until a robot does complex tasks.

For Davide Scaramuzza, director of the Robotics and Perception Group at the University of Zurich, the current problem is not in mechanics: “It has been around twenty years since we were able to show that machines are actually capable to carry out absolutely all the tasks of our daily life as humans, and sometimes even much faster than us, as long as we have programmed them for that.”

Interpret sensory data

Moreover, when it comes to flexibility and mobility, the new fully electric Atlas robot from Boston Dynamics, unveiled last April, can in fact make movements with its arms, legs, pelvis and torso that are completely impossible for a human body.

External content

This external content cannot be displayed because it may collect personal data. To view this content you must authorize the category Social networks.

Accept More information

“In fact, what this type of robot still largely lacks is the ability to interpret the data they acquire with their sensors from the external world,” continues Davide Scaramuzza. “For example, with the images from the cameras with which they are equipped, robots are still difficult to construct for themselves a 3D visual representation of their environment or to determine where the obstacles are.”

Another pitfall: disembodied systems like ChatGPT can train on a gigantic database, the entire Internet, to generate snippets of text probabilistically, with astonishing speed. But nothing like this in robotics: for this the robots would need to be able to practice on existing movement data samples, but these are largely lacking.

Olivier Dessibourg/asch

-

-

PREV Hervé Orama Barrere, new CEO of Centrale Danone
NEXT MA CNSS application… The National Social Security Fund launches a new version to improve the quality of its services – Telquel.ma