February 21, 2024


At Tesla’s 2023 shareholder meeting, Elon Musk showed off some new videos of the Optimus robot, which he believes will represent the vast majority of the company’s future value, and the development team’s approach to training artificial intelligence.

At least five Optimus robot prototypes have now been built and are able to walk, using motors, controllers and electronics designed and manufactured by Tesla. The general-purpose robot is learning to sense the world around it, map it and navigate using a system developed for the self-driving systems in the company’s electric vehicles.

“As fully autonomous driving gets closer to general-purpose real-world artificial intelligence, the same software could be transferred to humanoid robots,” Musk said. “Humans can obviously walk around with our arms and legs, but we can also drive cars, fly planes, steer boats, ride horses. If you have a generalized understanding, a generalized real-world AI, that’s us” Self-driving development, it’s transferable to basically anything. Optimus will use the same FSD computer as the car. “

“Very few people, even in the AI ​​community, appreciate Tesla’s capabilities in AI,” he added. “This is the most advanced real-world artificial intelligence to date. No one can match it. Reality has the greatest freedom.”

The articulation motor is powerful but sensitive enough to avoid breaking an accidental egg if necessary
The articulation motor is powerful but sensitive enough to avoid breaking an accidental egg if necessary

tesla

The video takes a quick look at some sort of leg design, showing off a degree of explosive jumping power from electric actuators, and the sensitivity of the system, with rubber feet stepping on the egg but stopping before breaking it.

It also showed some glimpses of the Optimus team training robots, using a similar monkey-see, monkey-do technique to the Sanctuary AI team we covered yesterday — but minus the telepresence component, obviously. Operators wearing motion-tracking suits and head-mounted cameras perform a variety of basic tasks that Tesla hopes to train the robots, and their movements are replicated by a virtual model of the robot that learns common movement patterns.

It’s worth noting that the operator is just doing the task himself, and the robot learns from the virtualization of the process, which is very different from Sanctuary’s method of letting the operator directly drive the robot through telepresence, which means that the operator is seeing only what he can see. What the robot can see, can only feel what the robot can feel — the robot’s own hands perform work under human control.

Human operators in motion capture suits perform basic tasks to train robots
Human operators in motion capture suits perform basic tasks to train robots

tesla

Intuitively, Sanctuary’s approach seemed better to me, providing better data for the AI ​​to learn from. This means that the human brain responds to the exact same stimuli that the robot encounters in the real world, e.g. Tesla’s apparently virtual approach can track a human picking up something, but it doesn’t provide any data from the robot . The robot itself is along with the process.

Either way, Optimus seems to be learning to pick up objects on its own—no small feat, as it requires the robot to see objects, map out their 3D shapes, figure out how best to approach them, and then make All the right choices for motor movement and angle changes to grab and lift them.

“At this point, Optimus Prime is not a very thoughtful person,” Musk said. “Optimus is still figuring out how to do basic things. It can’t boil eggs or anything. We need to get Optimus to a point where it’s reasonably agile and can do basic things. Our goal is to get it from doing boring Start with simple, repetitive or dangerous tasks—basically, jobs that people don’t want to do. That’s our goal, and I’m confident we’ll get there.”

Even the most basic real-world tasks are currently an excruciating challenge for AI-controlled systems
Even the most basic real-world tasks are currently an excruciating challenge for AI-controlled systems

tesla

“Then we had to figure out how to make it at scale,” he continued. “We’ve got to figure out how to keep it safe. The robot has to have a native way to shut it down…Safety is going to be extremely important, and I can’t stress that enough. But right now, it’s not in a place like Mars. The same level of intelligence to think about. Maybe one day.”

Musk reiterated his belief that Tesla robots will go far beyond the car business. “If you had a general-purpose humanoid robot, what would be the effective ratio of robots to humans? Because I think basically everyone would want one. Maybe people would want more than one. That would imply a real demand for something like Optimus — if it does work, it will — I don’t know, 10 billion units? That’s a crazy number. It could be 20 billion units. It’s a very large number, and that’s what I’m talking about, And that number far exceeds the number of cars. My prediction is that most of the long-term value of Tesla will be Optimus, and I am very confident in this prediction.”

No one accused Musk of lacking self-confidence, but it was fun to watch the Tesla robot take shape. We wonder how long it will take for this machine — or any of its competitors — to reach a point where it’s actually useful for a fairly wide range of tasks. I suspect, given the way the trajectory of AI generally develops, that once it reaches a point where it’s somewhat useful, it will evolve extremely rapidly from there.

Check out the video below.

(embed) https://www.youtube.com/watch?v=XiQkeWOFwmk (/embed)

Tesla Robot Update

source: tesla