Hot on the Internet! Is 2024 really the year of robots?

At the beginning of 2024, the robot industry dropped several heavy bombs and fired the first shot of the year.

Whether it is Stanford Mobile ALOHA, DeepMind's three major advances in robotics technology, or the Figure 01 robot, they have successively set off social media, pushing the exposure and discussion volume of "robots" on the Internet to new heights.

Does all this indicate that we have really entered the so-called "Year of Robots"?

Stanford's "all-round housekeeper" who can cook and clean

My dear family members, who understands the destructive power of roast chicken with scallops, shrimp with scrambled eggs, and lettuce with oyster sauce on Cantonese people!

What if the above dishes were made by robots, how would you respond?

The robot, holding a pan in one hand and a spatula in the other, fried the boneless chicken thighs until both sides were golden brown, then added scallops for flavor and seasoning, simmered until the sauce was reduced, and finally sprinkled with chopped green onions. A tempting scallop-roasted chicken was ready! (When will I have a phone that can smell? I'm waiting online)

It can not only cook, but also "clean up the battlefield", wash pots and tables, and organize cupboards, all with ease. Its business scope is not limited to the three inches of the kitchen, and "everything is alive" wherever it can see.

Watering the flowers and playing with cats, washing clothes and folding quilts...all of these are no problem, and it can even help with shaving!

“Isn’t this the nanny robot we’ve always dreamed of?”

For a while, many netizens were amazed and said, "The robot with living eyes is finally here!" (Let the parents see that it is not only inferior to the barbecued pork, but this time it can't even compare to the robot)

This robot, named Mobile ALOHA, comes from a research team at Stanford University consisting of doctoral students Zipeng Fu, Tony Z. Zha and assistant professor Chelsea Finn.

The research paper states that Mobile ALOHA is a low-cost mobile operating platform consisting of a mobile base, a two-handed operating system, sensors and cameras that can collect and train robot motion control data.

The research team found that through supervised learning methods, Mobile ALOHA can learn mobile manipulation tasks in complex environments and complete operations such as calling elevators and opening cabinet doors, with a success rate of up to 90%, after watching humans demonstrate the same task 50 times.

Mobile ALOHA adopts a fully modular and open source design, with an overall hardware cost of only $32,000 (about RMB 220,000), of which the mobile platform costs only $7,000 and is equipped with a 14kg battery to power the robot, allowing it to move autonomously; and in terms of the "brain", the Intel RTX3070Ti chip is enough to support its computing. This has a significant price advantage over many professional robot platforms, and the research threshold is greatly reduced.

After the video went viral, the research team "contradicted themselves" by releasing a "crash" video to clarify that Mobile ALOHA is not omniscient and omnipotent, and that many of its capabilities still require remote human control. Many of the "extreme operations" in the video also use a hybrid mode.

After Zhao released the highlights of his failures, he encouraged everyone to pay attention to the original paper and said, "This may be my favorite video so far." Some netizens also agreed, saying, "It's cute!" (Don't worry about being replaced for now)

When the Mobile ALOHA video came out, it really shocked many people, making them sigh that the automation of robots has developed to such a sophisticated level, but its "hybrid" was actually expected.

Firstly, according to existing technology, it will take some time for a robot to complete such complex and fully autonomous operations; secondly, the creative team did not intend to hide it at all. As long as you pay close attention, you will find that manual control appears in many shots, and there is also an explanation of the robot driver at the end of the video.

But it is obvious that most people are attracted by the robot's showmanship and do not pay attention to the final details, let alone read the project website and original paper.

Nvidia scientist Jim Fan also posted on the platform: We are still a long way from having fully autonomous robot chefs or maids, but I am still excited about this new research!

Google DeepMind's three major breakthroughs

On the same day that Mobile ALOHA was released, DeepMind launched three major research advances in terms of data collection, decision-making speed, and generalization capability.

Picking up fruits, taking out Coke from the drawer, placing toothbrushes... The robot is able to perform these operations using an automated data collection system AutoRT and a new model SARA-RT that accelerates decision-making.

The researchers spent seven months using the AutoRT system to control robots to complete tasks. They have collected diverse data covering 7,000 trials and 6,650 unique tasks. They can control up to 52 robots at the same time, increasing robot operation speed by 14% and accuracy by 10.6%.

In addition, there is a new framework RT-Trajectory that focuses on generalization capabilities. It helps the robot to deeply understand how to complete a task by explaining its specific actions, rather than simply matching instructions and actions. This allows the robot to have a success rate of up to 63% when faced with 41 tasks it has never seen before.

While launching a series of important system advances, the Google DeepMind team also drafted the world's first constitution for intelligent robots .

It is inspired by the "Three Laws of Robotics" by science fiction writer Asimov, the main core of which is to ensure that robots do not harm humans. DeepMind has set up guardrails for the LLM model to ensure that the task suggestions it generates do not involve unsafe content such as humans, animals, and sharp objects. At the same time, it has programmed restrictions on the robot's joints and added physical switches for control.

Previous discussions surrounding AI ethics have mostly focused on issues such as whether the algorithms themselves are biased, and rarely on what ethical standards AI, especially intelligent robots with physical form, should follow.

The first robot constitution embodies the idea that technology should serve humanity, rather than simply pursuing functionality and efficiency. Although as a first step, this robot constitution is still relatively simple and principled, how to make it have a more rigorous logical system and how to cover various situations and ethics still requires extensive discussion and continuous improvement.

But the history of the progress of human civilization is a process of continuous learning, exploration, making mistakes and corrections. As we stand at a new stage of technological development, since we have taken the first step, its influence will by no means be limited to the technical level.

Figure 01 Learn how to make coffee by watching the video

This Figure 01 robot comes from Figure, a company in Silicon Valley that has great money-making power.

Figure, which claims to be the world's first commercially viable general-purpose robot, completed a total of US$79 million (approximately RMB 5.66 million) in financing last year, and its shareholders include star companies such as NVIDIA.

Following its debut last October, Figure 01 is no longer “staggering” around, it’s brewing coffee!

Figure 01 is equipped with an end-to-end AI system. By watching 10 hours of videos of people making coffee, it can completely complete the steps of opening the lid, putting in coffee, pressing the brewing button, etc. completely autonomously. If it encounters problems in the middle, it can also self-debug and optimize.

The company's founder said that the video speed has not been accelerated in any way, so it seems that the operating speed and smoothness of Figure 01 are still very impressive, even surprising.

If it has strong generalization capabilities in the future, Figure 01 can achieve rapid transfer learning through video demonstrations, allowing robots to break away from traditional programming models and turn to big data and AI to achieve autonomous control, greatly optimizing human-computer interaction.

Conclusion

As the popularity of AI technology continues to rise, humanoid robots and general-purpose robots have become a new trend in the industry. Not only have many high-level players entered the market, the country has also introduced policies to strongly support them. They have become the new generation of "traffic kings" in the machine industry.

According to Goldman Sachs' forecast, the robot market space will reach at least US$6 billion in the next 10 to 15 years. In the best case scenario, it is expected to reach US$154 billion in 2035.

However, even though there is great potential and we are trying hard to catch up, with today's technology and market, robots are still more than one or two steps away from the goal of "entering thousands of households."

However, the latest achievements of Mobile ALOHA, DeepMind and Figure have also boosted confidence in the robotics industry, making the "Year of Robotics" more promising. When these advanced technologies are truly implemented from the laboratory to the real world and integrated into our daily lives, perhaps we will usher in the true "Year of Robotics".