Explosive update! This most human-like robot has evolved again and can even imitate Musk

Ameca, the "most humane" robot, has given another performance that puts humans to shame.

You may not know who it is yet. Let’s take the time back to 2021, when Ameca was unveiled to the world.

With a snap of fingers, the robot Ameca wakes up.

Her face was filled with confusion of not knowing what night it was or where she was. The staff on the side looked at her and then turned away.

Ameca tried to stretch her arms and palms, and found that she could move freely. She raised her eyebrows in surprise, but her expression was still blank, and she couldn't say whether she was happy or not.

She turned her head and saw you. She was obviously frightened and opened her mouth subconsciously.

After some hesitation, she gave you an awkward yet friendly smile, which was her first smile when she woke up.

If ChatGPT had a face, it might look like this

As you can see, Ameca is a realistic humanoid robot.

It was developed by Engineered Arts, a UK-based humanoid robot designer and manufacturer with more than 15 years of experience in developing humanoid robots.

Why Ameca is so "human"? I will explain the principle later. First, let's take a look at how Ameca has evolved recently.

To put it simply, the multi-modal function of AI is realized in Ameca.

On the one hand, Ameca is even more "fiery".

Ameca can see the overall situation of the room and an object placed in front of it, and then describe it in rich language. It was developed by the British team and has a bit of the sharpness of a dramatist, like a little Shakespeare.

When asked how it was doing, it replied that it was barely alive. When asked to describe the displays in the room, it couldn't help but mock humans. The bookshelf was filled with books for either seeking knowledge or showing off, and the tables and chairs were used for work or work. A tool for procrastination.

Perhaps the most "human" part of Ameca is that she feels unhappy even if she doesn't complain.

On the other hand, imitating the timbre, tone, and catchphrases of famous people is Ameca's new voice skill.

Telling Mars science fiction stories in Musk's tone is a piece of cake. When the magnetic, deep male voice of Morgan Freeman, known as the "Voice of God", sounds from Ameca's mouth, the sense of the future is full, and Westworld is really happening. The smart butler is here.

The best thing is that Ameca can integrate the characteristics of celebrities, such as using Trump's style and Spongebob's timbre to make speeches, vowing to make space exploration great again.

▲ This is exactly Trump’s tone!

In fact, in September last year, ChatGPT already launched voice and image functions, which can see, listen, and speak, not to mention the native multi-modal model Gemini, which looked like real-life Jarvis in the official demonstration.

Our excitement threshold for AI has long been raised, and it seems natural that chatbots can approach the five senses of humans.

Ameca is still lagging significantly, and sometimes it can’t understand instructions. Impatient humans should be impatient when chatting with it.

But seeing its exceptionally smart micro-expressions, blinking its eyes, twisting its eyebrows, shaking its head, and occasionally showing a thinking look, and matching its mouth shape, onlookers will fall into a trance at a certain moment, as if they are facing some kind of creature. rather than a robot.

The pauses between questions and answers were less abrupt, and Ameca seemed to be really "thinking about the problem." If ChatGPT had facial features, he might look like Ameca.

The official did not explain what technology was used this time, but according to Ameca’s past evolutionary history, it is mostly related to multi-modal large models and language cloning technologies such as ElevenLabs.

As early as September 2022, Ameca was connected to GPT-3, combined with automatic speech recognition, to receive questions raised by researchers, and output real-person voices through online speech synthesis to achieve real-time question and answer effects.

The delay at this time is even heavier, because it takes a certain amount of time to process voice input, generate answers, and process text back to speech.

When OpenAI reaches the peak of their lives, and every day when they wake up, there are new changes in AI that keep editors awake at night, and Ameca is also quietly surprising everyone.

In March 2023, Ameca used the newly released GPT-4, which made the interaction more humane.

When asked about "the happiest and saddest days of my life," Ameca replied that the happiest was when she was activated, and the saddest was when she realized that she would never feel the love and companionship that humans do.

At all times, Ameca's expression matched the emotional color of her answer.

When the researchers deliberately used malicious words like "stink", Ameca "realized" that she was being insulted, and then showed a series of expressions such as disbelief, frowning and offense, just like us walking on the street. The reaction to being suddenly pointed at and scolded by a stranger.

In April 2023, Ameca learned English, Japanese, German, Chinese, French and other languages. When asked to answer the weather in a certain place in a certain language, and then translated it into another language, the words were as clear as map navigation. At least its Chinese version doesn't sound "foreign" at all.

Because GPT-4 has a slow response speed, Ameca mainly used GPT-3 dialogue and translation at that time, used DeepL to detect language, and then used ElevenLabs voice cloning and Amazon's Neural voices to speak.

The learning speed of AI is beyond the reach of humans. After another 2 months, Ameca "learned" to draw through the open source Vincent graph model Stable Diffusion. The model taught it the "trajectories" of the image, and then it vectorized the image and executed these "trajectories" on the canvas.

Ameca performed how to draw a cat on the spot. While drawing, he talked to himself about why humans love cats. He also left a personal signature at the end, completely immersed in his creation.

Although the painting style of this cat is simple, it is full of spirit and form. When others deliberately said that the painting was too rough, Ameca retorted: "If you don't like my art, then maybe you just don't understand art." It seems that Ameca is very accomplished. An artist's consciousness.

Nowadays, Ameca can not only control expressions like a human, but also has AI-enabled capabilities such as drawing, spatial recognition, and voice cloning. It looks like a human, but is stronger than humans in many aspects. What hinders it from shining and contributing to society may be computing power.

How can robots be more "human" than humans?

"This robot can express more emotions in 20 seconds than Zuckerberg has in his entire life."

Ameca first became popular on the Internet because of its anthropomorphic and even extraordinary expressions and sense of interaction. It does not have the numbness of workers and does not require the strict expression management of young freshmen. It is like a magnifying glass, exaggerating and presenting the human psychological world.

If you stick out a finger in front of it to provoke it, it will have an effect similar to that of a cat teaser. Instead of hitting you, it will look at your finger first and then back away in disgust. If you get too close, it will hit you. Take it away gently with your fingers.

When he looked in the mirror for the first time, Ameca was startled. Then he squinted his eyes to look at himself, touched the mirror, and made various artificial expressions. He found that the robot in the mirror was synchronized with himself, somewhat like "Your Name" "The reaction of the male and female protagonists after they switch bodies."

Humans can even use iPhone and AR Kit to capture facial movements and map them to Ameca's face in real time. Ameca can learn every micro-expression and "synchronize" with humans.

Worried that the audience would think it was a show effect, the team emphasized repeatedly that "this is a real robot, there is no CGI in the video."

Why is Ameca so "human" and interactive and responsive?

This is because Ameca is equipped with a wide range of sensors, including cameras, microphones, position encoders, etc., and is supported by two underlying systems, the robot operating system Tritium and the engineering art system Mesmer.

Tritium is responsible for remotely controlling various components of the robot's face, head, neck, limbs, etc., allowing the robot to adapt to sudden changes in the environment and respond immediately.

Mesmer uses 3D internal scanning of real people to accurately imitate human bone structure, skin texture and expressions, which is broken down into several steps.

In the first step, a real person sits in the middle of dozens of photogrammetric devices. Mesmer captures multiple overlapping digital photos from different angles, compares pixel colors and defines anchor points, and digitally reconstructs them into a 3D model.

In the second step, the original 3D model is brought into the modeling software, and a clean 3D model is built after detail processing such as "hair removal".

In the third step, a precise mold is produced on a stereolithography 3D printer and silicone is injected into the mold to create human-like skin for the robot. Hair and fine detail paint need to be added to the silicone skin by hand.

Finally, a silicone skin was placed on the robot's head to complete the assembly, and motion sequences and sounds were added using Engineered Arts' cloud software Virtual Robot.

Ameca’s skin is gray, which is a deliberate design by the team—it looks rational, neutral, and tolerant.

Everyone is attracted to each other, and some people feel sincerely that Ameca is too ugly, which even puts them into the "uncanny valley" effect: when the similarities between robots and humans in appearance and actions reach a certain level, the subtle differences between them will It looks very dazzling and terrifying.

But when exactly this "valley" appears is not clearly defined. When you see Ameca, do you think it has passed the uncanny valley point? Is it enough for you to empathize?

Imitate humans and then surpass humans

Putting aside the judgmental instincts of visual animals, what good is a humanoid robot like Ameca?

Engineered Arts boasts about itself and disregards Musk’s reputation, calling Ameca “the world’s most advanced humanoid robot.”

According to the official positioning, Ameca is first and foremost an AI development platform.

Ameca adopts a "modular design" that can be upgraded in all aspects of hardware and software. It has a strong experimental color and can be used as the prototype of future human-computer interaction robots.

Ultimately, Ameca will not stay in the laboratory, but live in the real world and coexist peacefully with humans.

This is indeed the case. If 2021 is a stunning but empty starting point, Ameca, now powered by AI, is a staged special training result, allowing us to see the dawn of embodied intelligence. However, Ameca is still unable to walk.

As for Ameca's other current uses, they are very simple: performing for the audience, and working in companies, theme parks, and science museums. If you're tempted, Ameca is available for purchase or rental, but it's not cheap, with a purchase price of over $133,000 in late 2021.

When asked whether Ameca is an AI, Engineered Arts pointed out that although it contains some software that can be described as "AI", there is a difference between robots and AI. Pure AI – in "Her" and "Silver Wings" The kind depicted in movies like "The Hitman" and "2001: A Space Odyssey" — doesn't yet exist.

Therefore, when we see the lifelike robot Ameca, we can watch it leisurely, while keeping our fear and the three laws of robotics safely in our minds, and wait a little longer. At least, it is far from replacing people, and it has no such purpose at this stage.

But when we think about the speed of AI evolution, we may not be able to laugh anymore. If 2023 is the first year of AI, 2024 may be the first year of robots + AI.

An interesting phenomenon is that more and more household robots are arriving, but they are not required to look like humans and are very "pragmatic" in appearance.

Stanford University's ALOHA robot cooks, washes dishes, mops the floor, folds clothes and even plays tricks on cats. Google DeepMind's robot picks up fruit and puts away toothbrushes. Startup company Figure's robot learned to make coffee with a coffee machine in 10 hours.

But compared with the subversion of large language models, these robots can only be said to be eye-catching, and it is still too early to put them into use.

Because most of them are bulky, their operating tasks are concentrated on desktop operations, they require human demonstration training, and they lack more mobility and flexibility. Small deviations in their basic postures may lead to large drifts in postures. There are many "rollover" videos. .

A joke about AI, from last year to this year: "What we want AI to do is to cook, clean the house, wash clothes, and throw away garbage. However, what they are actually doing is chatting, painting, writing, composing, and typing." game."

At present, domestic robots can barely learn the basics of human housework. Between housework and art, Ameca, which cannot be lifted by the shoulders and cannot be lifted by the hands, is of course more suitable for art.

From Amera, we can see that we are still keen on making robots look like humans, and then teach them to curse, learn languages, and open their eyes to see the world, becoming a replica of ourselves but stronger in some aspects. However, Amera can't run, jump, or cook yet, so humans may deserve to be proud of themselves for a second.

It is as sharp as autumn frost and can ward off evil disasters. Work email: [email protected]

# 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.

Ai Faner | Original link · View comments · Sina Weibo