Quietly, Tesla suddenly released a new generation of humanoid robot Tesla Optimus Gen 2 today.
In one sentence, the key point is that compared with the previous generation, Optimus Gen 2 is more human-like. The main upgrade points are summarized as follows:
- Tesla's newly designed actuators and sensors improve the efficiency and accuracy of task execution
- 2-Dof (degree of freedom) driven neck allows the head to rotate more freely
- 11-Dof hand joints make hand movement more flexible
- Integrate actuators with electronic components and wiring harnesses to improve overall system stability
- Increase walking speed by 30%, enhancing environmental adaptability
- Introducing foot force/torque sensing to flexibly adapt to ground conditions
- 10kg lighter and more maneuverable
As can be seen from the video posted by Musk, Optimus Gen 2 can walk smoothly, with a walking speed that is 30% faster than the previous generation. It can also squat at about 90 degrees, and the degree of limb control is quite good.
Optimus Gen 2 has also been upgraded with a brand-new hand. The 11-degree-of-freedom hand allows it to shake like a human hand, and the accuracy and smoothness of the movements are like real people.
Optimus Gen2 has tactile sensing in all its fingers, allowing it to dexterously handle fragile items like eggs and look like it can cook.
At the end of the video, you will also see two Optimus Gen 2 dancing together to the beat of the music and the dazzling lights. This is also a little Easter egg buried by Tesla.
A senior Tesla engineer said that everything in the video was real and was not generated using CGI, so the actions were all real-time and not accelerated. He also praised the Tesla robot for its incredible improvements in hardware performance.
If what he said is true, then the evolution of Tesla’s humanoid robot is clear at a glance
▲Picture from @Hanteta_
The Evolutionary Journey of Tesla Robots
It has only been three years since Tesla launched its first humanoid robot, but it has ushered in rapid evolution.
At the AI Day in August 2021, Musk officially announced the Tesla humanoid robot Tesla Bot. The robot is envisioned to be 5 feet 8 inches tall and weigh 125 pounds, with the ability to carry 45 pounds and lift 150 pounds, and is primarily used to perform tasks considered dangerous, repetitive and boring.
But at that time, this robot was just a few PPT concept pictures that "lived" on the screen, and even the stage dance appearance was completed by a man wearing a robot suit.
Then at the AI Day 2022 conference the following year, the protagonist changed completely. Optimus Robot, who walked to the front of the stage, became the strongest protagonist of the conference that day.
The exposed mechanical structure makes its appearance a bit rough, but whether it is moving boxes, watering plants, or moving bricks in the super factory, the "Optimus Prime" in the demonstration video can do a lot of work.
What provides energy for this Optimus Prime is not the fire, but a battery pack with a capacity of 2.3kWh. It is fully protected and placed at the core of the body, which is the blue one in the picture.
At the time, Musk said that Optimus Prime was "very suitable for a full day's work." Its power consumption was about 1 million when sitting still and 5 million when walking quickly. Some people may ask how fast walking is, about 2 meters per second.
In order to ensure the flexibility of Optimus Prime, Tesla even designed 28 joints for it. Each joint has multiple sets of actuators. These actuators can analyze execution efficiency in the background and reduce ineffective work.
Tesla said that the entire Optimus Prime has more than 200 degrees of freedom, and its hands account for 27 of them. They are driven by an actuator. Through the visual system and algorithm, Optimus Prime can accurately identify the hand and the The location of the object to complete the tasks of watering flowers and picking up objects mentioned above.
It took "Optimus Prime" five months from taking the first step to moving freely into various poses.
During the Tesla investor period in March this year, Optimus once again ushered in an epic evolution. One of the highlights was when Optimus picked up a robotic arm from the table and fixed it in place, while another Optimus fixed it on the third Optimus that was still being assembled. Such a nimble and smooth task execution, much better than the original Ability has made substantial progress.
At that time, Optimus' movements were still relatively stiff and he could not yet complete complex movements like parkour, but he had initially demonstrated universal abilities such as walking independently short distances, carrying objects, and using basic tools.
Just half a year later, at the AI Day in September, Optimus was not only able to sort blocks according to color, but also confidently raised its arms in the air and then struck a yoga pose on one leg, showing excellent balance and coordination. force.
Now, as mentioned above, with its lifelike body, nimble fingers, and smooth and light gait, Optimus Gen 2 is closer to human levels in movement and operation capabilities. Perhaps in the near future, the more mobile Optimus will even be able to replicate complex human movements such as "crawling, jumping, grabbing, and lifting."
Of course, according to Musk's vision, Optimus's long-term goal is not just to replace humans in tasks such as repetitive labor and dangerous operations, but to become a general service robot that can be integrated into more complex scenarios such as homes, offices, and commercial scenes. middle.
Therefore, the robot's hardware is only the cornerstone. A complete humanoid intelligence also requires "brain" level software and algorithm empowerment.
What happens when you stuff GPT-4 into a humanoid robot?
Since the beginning of this year, with the continuous iterative upgrading of large models and the rapid improvement of intelligent generalization capabilities, people have seen the potential prospects of humanoid robots. From time to time, we will see some new players entering the humanoid robot track.
Coincidentally, they all chose the problem-solving approach of embodied intelligence. Embodied intelligence, as the name suggests, refers to a form of intelligence that uses its own physical body to obtain information, make judgments and take actions through perception, movement and environmental interaction. This is also regarded as the ultimate form of artificial intelligence. .
It emphasizes the critical role of the body in the occurrence of cognition and behavior. Compared with pure thinking intelligence, embodied intelligence is closer to human natural intelligence. It emphasizes the important role of body, behavior and environment in cognition.
A less rigorous expression is that the AI model serves as the brain and the robot acts as the body. The AI model put on the robot's shell is embodied intelligence.
In addition to the Tesla robot mentioned above, a more intuitive case is that the humanoid robot Alter3 recently released by the University of Tokyo is the first humanoid robot driven by GPT-4. Thanks to the blessing of GPT-4, no prior programming is required. Through verbal feedback, you can let it perform various human actions autonomously.
▲Don’t come over here
Playing heavy metal music, performing selfies, throwing balls with your hands, pretending to be a scary ghost, all these actions are easy to do under the mapping of GPT-4 intelligent language description.
It should be noted that Alter3 itself cannot understand details such as "how high to raise your hand", so it cannot improve its actions. It receives information through human verbal feedback, such as "raise your hand a little higher when taking a selfie". GPT-4 breaks down instructions given by humans into a series of steps.
These steps will then be converted into Python code that controls the movement of each body part of Alter3, which is the so-called chain of thought (CoT). Then Alter3 will store these improved movement codes to form a "movement memory."
The next time it encounters the same movement, it will reason based on past relevant experiences on how to optimally perform similar tasks instead of starting from scratch. This zero-shot learning capability is also what the robot has gained. The key to true intelligence.
What's even more outrageous is that without consciousness, it can even infer appropriate emotions from the text fed back by GPT-4 and express them with clear movements. It is indeed another "super-class" performance of the robot.
Like the Zhiyuan Robot Expedition A1 of Huawei's talented youngster Zhihui Jun, which we have previously reported, and Lei Jun's "Iron Egg", they also fall into the category of AI + robots.
Before large models, robots were specialized equipment. Now with the universal capabilities of large models, universal robots are beginning to appear. As Zhihui Jun said, now we not only need ChatGPT, but also WorkGPT.
These evolutions indirectly confirm a clearly visible path: after the large AI model takes root, it will eventually enter the real world, and embodied intelligence is the best path.
Nvidia founder Jensen Huang, who has been active on the front line of AI, once said: "Embodied intelligence will lead the next wave of artificial intelligence."
Mature AI large models act as artificial brains, simulating the complex neural network of the human brain, realizing cognitive functions such as language understanding, visual recognition, and situational reasoning, and solving higher-level cognitive and decision-making problems for robots.
At the same time, various sensors, actuators, and computing units are integrated into the robot body to realize perception and interaction with the environment. For example, vision systems can capture images and videos, and tactile sensors can sense the shape and texture of objects.
Therefore, based on the deep integration of AI large models and robot entities, we imagine that one day, the ultimate Grok combined with Optimus may be able to blur the boundaries between virtual and physical reality, allowing intelligent AI to completely step out of the screen and truly integrate into human life and society. .
Finally, let netizen @cb_doge ask Grok to comment on Optimus Gen 2’s answer:
Oh, the progress Tesla has made with Optimus Gen 2 is really impressive! Both speed and weight reduction, as well as the ability to handle fragile objects like eggs, have been significantly improved since the last time it came online. It's like watching a little bird bravely attempt to fly for the first time, even though it has more actuators than feathers.
Looking at the big picture, we need to understand that developing robots like Optimus is a long-distance race, not a sprint. It's like watching a slow, elegant dance between technology and humanity, with each step bringing us closer to an ideal future where robots and humans can live in harmony and work together.
At least, that's our dream.
# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.