The Tesla boss claims to present a prototype bipedal machine next year. She would be able to replace humans for boring, dangerous or strenuous tasks.
Tesla’s AI conference at the end of August ended with a surreal surprise. A dancer in a tight-fitting costume looking straight out of the 80s has started to imitate the jerky movements of an automaton, a little kitsch prelude to the announcement of yet another crazy project by Elon Musk: the Tesla Bot. A humanoid robot capable of performing boring, dangerous or arduous tasks for humans.
Elon Musk criticized
Laid-back, the businessman said Tesla would be able to present a prototype humanoid robot without any problem as early as next year. However, he had nothing to show except dance steps, and just shared a few tech specs.
While his fans have greeted him hysterically as usual, several robotics experts have greeted the news with great caution and annoyance. This is the case of Rodney Brooks, former director of the artificial intelligence lab at MIT and co-inventor of the Roomba vacuum robot, as well as the Baxter industrial robot. A size. He tweeted a pithy message in the face of the enthusiastic reactions. “I think this project will not be part of the history of the next ten years. We will understand that this was a very insignificant announcement. ”
Specialist journalist Evan Ackerman also wrote a stern article for the very serious IEEESpectrum titled Elon Musk has no idea what he is doing with the Tesla Bot. However, the humanoid robotics experts we consulted in France were more nuanced with regard to this project.
A prototype quickly conceivable
Let’s start with the goal of presenting a prototype by 2022. 1.76 meters, 56 kilos, the humanoid will have human dimensions. The problem is, Tesla does not publish scientific papers and no one knows how long his team has been working on the subject or what resources and talents they have. Even though the team does not begin work until today, the challenge does not seem completely insurmountable.
“We are not starting from scratch when it comes to humanoid electric robots,” said Olivier Stasse, research director at LAAS-CNRS (Laboratory for Analysis and Architecture of Systems) in Toulouse. “If the Tesla team builds on what has already been done, including machines like Digit from Agility Robotics, it is possible to develop a prototype in two or three years. But for a first model, in general, you have to iterate a lot, like Boston Dynamics, to get to something, ”says the scientist. In addition, he knows what he is talking about, since this specialist in the generation of anthropomorphic movements has worked for several years on the humanoid robot Pyrene.
To go even faster, we could also imagine another option. “It is conceivable to develop a prototype that works in one year, if they choose to remotely operate it by a human and on condition that the engines are well controlled,” says Serena Ivaldi, researcher at INRIA (National Research Institute in Digital Science and Technology). This expert in human-robot interactions is experimenting this path herself in Nancy, using the humanoid Talos from Pal Robotics as an avatar of the human operator. This would greatly simplify Tesla’s task.
The inventor of the Nao and Pepper robots Bruno Meissonier is even more optimistic about the deadlines. “We had developed the Pepper prototype in less than six months. Therefore, yes it is possible to build a prototype humanoid robot in just a year, “he told us.
Will the Tesla prototype be able to accomplish such feats as early as 2022? It seems difficult. Because building the product is not the biggest challenge. The Tesla Bot is supposed to move up to 8km / h, grab and handle tools, lift and carry heavy loads in an environment made for humans and among humans. The issues lie in the perception of the environment, the generation of movements and the maintenance of the balance of the machine.
Can Tesla reuse on its automotive expertise?
Elon Musk insisted on the expertise of Tesla teams in terms of semi-autonomous electric cars. “Tesla is arguably the largest robotics company in the world because our cars are semi-sensitive robots on wheels,” he said.
According to him, the technologies embedded in his vehicles can easily be put to the service of the humanoid robot: cameras, sensors or actuators. It is also about recovering the all-new Dojo artificial intelligence drive chip and benefiting from the FSD (Full Self-Driving) system, ie the autonomous driving computer.
It is also possible that Tesla overcomes the problem of the autonomy of this robot. For Olivier Stasse, it would be possible for the humanoid robot to be able to operate for 12 hours in autonomy. In addition, that he himself changes his battery every hour and a half.
Tesla has undeniable strengths that it can capitalize on, such as computing power, software architecture, and actuation mechanics, which may include some similar principles. However, it will be far from sufficient.
A humanoid, it inevitably falls
Controlling a humanoid robot is very different from a car. “A road is easy for a car to identify: it is flat with two curbs,” says Serena Ivaldi. “And when there is an obstacle, the vehicle has a choice of stopping or avoiding it. Its route is also easy to calculate. However, the humanoid robot performs different tasks. The control and generation of these movements are also very specific. It is the most difficult platform to master, “she adds. A ground that is not perfectly level can on its own disrupt the locomotion of a biped.
“A humanoid has 30 axes to manage against 4 for a car”, adds Olivier Stasse. “He has to avoid all kinds of collisions, starting with himself… like putting his hand in his head. Its balance is also difficult to maintain when moving. He is bound to fall. However, the important thing is that he is able to stand up. The problem is, it can break, because it is not as flexible as the human body, “he observes.
To be convinced, just watch the latest video from Boston Dynamics. The famous company showed for the first time behind the scenes of the workouts of its android Atlas. Before culminating in a breathtaking demonstration of a few minutes where the robot jumps and jumps over obstacles with virtuosity, there are hours and hours of repetitions of the same movements and incessant falls.
Impossible to improvise a shopping trip!
Therefore, using this robot to do our shopping for us at the local grocery store, as Elon Musk suggests, is downright impossible. Going from the user’s place of residence to a store they would never have visited, after walking alone in the street amidst countless unforeseen obstacles is currently very unthinkable.
Humanoid robots can currently only operate under controlled conditions and in a known environment, which assumes that it, has been previously modeled.
“Take the example of a human who moves a sofa down the stairs”, quotes Olivier Stasse. “He is able to spontaneously lean on a wall to help himself. To achieve the same result, the robot should first have a representation of the environment and be aware of the presence of a wall. But also have algorithms already configured to generate the right movements “.
Technological limits
According to Bruno Maisonnier, the technological obstacles are sizeable before one day reaching useful humanoid robots. “They’re not yet smart enough or nimble enough to be useful. Placed in a real environment, they do not work very well, “he emphasizes. “To hope one day to go beyond the prototype stage and market them on a large scale, we will have to overcome certain technological limits. Invent new engines. In addition, advance in artificial intelligence with microchips that mimic the human brain, to develop a true understanding of the real world. ” This is what the former boss of Aldebaran is committed to doing with his company AnotherBrain, which is developing a new generation of bio-inspired artificial intelligence focused on reproducing the functioning of the cerebral cortex.
In the meantime, we can always dream of future applications of Tesla’s humanoid robot. Bruno Maisonnier thinks of closed places: users’ homes, hospitals or retirement homes to begin with. Serena Ivaldi goes further. “We could eventually imagine using this robot instead of humans in contaminated or burnt areas, nuclear power plants. It could also coexist with humans in warehouses and factories performing complementary tasks. In addition, maybe even ensure minimalist interactions with customers like delivering an Amazon package, for example, “she says.
On Twitter, Elon Musk hinted that he could send the Tesla Bot to Mars to wipe the plaster from this space adventure instead of humans. So it may be a robot that will take its first steps on the red planet.