Integration of DePIN and embodied intelligence: technical challenges and future prospects

Introduction: On February 27, Messenger Bodcast hosted the “Decentralized Physical AI Building” and called Michael Chu, co -founder of the Frodobot Laboratory. They talked about the challenges and opportunities of the decentralized material infrastructure networks (DePin) in the field of robots. Although this field is still in its cradle, it has great potential and may completely change the way the artificial intelligence robots work in the real world. However, unlike traditional artificial intelligence that depends on large amounts of internet data, DePin Robotic AI faces more complicated problems, such as data collection, hardware restrictions, evaluation suffering, and sustainability of economic models.

In today’s article, we will dismantle the main points in this discussion and see the problems faced by DePin Robotics, what are the main obstacles to decentralized robots, and why DePin has advantages on central methods. Finally, we will also explore the future of the Depin and know if we are about to enter the “Chatgpt” of Robotics DePin.

Where is the bottle neck from smart robots?

When Michael Cho began working on Frodobot for the first time, the largest headache was the cost of robotics technology. The prices of commercial robots in the market were ridiculously high, making it difficult to promote artificial intelligence applications in the real world. The first solution was to build a low -cost independent robot costing only $ 500, and it intends to win by being cheaper than most current projects.

But when he and his team were deepened in research and development, Michael realized that the cost was not the real bottleneck. The challenges of decentralized material infrastructure networks (DePin) in robots are much more complicated than “expensive or not.” As Frodobotlab continues to progress, multiple bottlenecks appeared in DePin Robotics gradually. To achieve widely publishing, the following bottlenecks must be overcome.

Pubble neck 1: data

Unlike the large “online” artificial intelligence models that are trained in large quantities of Internet data, the embodiment AI needs to interact with the real world to develop intelligence. The problem is that the world currently does not have a large basis, and there is no consensus on how to collect this data. Data collection can be divided into the incarnate AI into the following three categories:

▎ The first category is human operation data, which is the data created when humans manually control robots. This type of data is of high quality and can capture video flows and motion marks – that is, what humans see and how they interact accordingly. This is the most effective way to train artificial intelligence on the tradition of human behavior, but the disadvantage is that it is expensive and intense.

▎ The second type is artificial data (simulated data), which is useful for training robots to move in complex terrain, such as robots training on walking on rough land, which is very useful for some specialized fields. But for some tasks with a variety of changes, such as cooking, the simulation environment is not very good. We can imagine that robot training is on egg frying: the type of pan, oil temperature and minor changes in the room conditions will affect the results, and it is difficult for a virtual environment to cover all scenarios.

▎ The third category is the video learning, which is to allow the artificial intelligence model to learn by monitoring videos of the real world. Although this method has potential, it lacks the actual actual interactive comments required for intelligence.

Pubble 2: Self -Government level

Michael stated that when Frodobot was first tested in the real world, he used mainly the robot for delivery in the last tendency. From the point of view of the data, the results were actually good – the robot successfully completed 90 % of the delivery tasks. But in real life, the average failure of 10 % is unacceptable. Robot cannot fail once in every ten deliveries at all. Just like automated driving technology, driving driving can have a record number of 10,000 successful leaderships, but the only failure is sufficient to defeat trading consumer confidence.

Therefore, for the robots to be really useful, the success rate must be close to 99.99 % or higher. But the problem is that every 0.001 % increase in accuracy requires time and a miserable effort. Many people reduce the difficulty of this last step.

Michael recalled that when he sat in the initial model of the self -driving car in 2015, he felt that full independent driving was just around the corner. Ten years later, we are still discussing when we can achieve a complete independence of the 5 level. The advancement of robots is not written, but it is worse – with every step forward, the difficulty increases significantly. This last accuracy 1 % may take years or even contracts to achieve them.

Porting 3: Devices: Amnesty International alone cannot solve the robot problem

In taking a step back, even if the artificial intelligence models are strong, current robot devices are not ready to achieve true independence. For example, the most easily ignored problem is the lack of touch sensors – the best current technology, such as the Meta AI search, is far from the sensitivity of the tips of human fingers. Humans interact with the world through vision and touch, while robots know almost anything about texture, grip and pressure.

There is also a blockage problem – when an object is partially banned, it is difficult for the robot to recognize and interact with it, while humans can understand an intuitive object even if they cannot see its full appearance.

In addition to perception issues, the automatic engines themselves have defects. Most of the human robots put the operating directly on the joints, making them huge and dangerous. On the other hand, the structure of human string allows smoother and safer movements. This is why the current human robots look harsh and flexible. Companies like Apptronik develops more vital operating designs, but these innovations will take a long time to mature.

Porting 4: Why are the expansion of devices very difficult?

Unlike traditional artificial intelligence models that depend only on computing power, the achievement of smart robot technology requires the spread of physical devices in the real world. This represents the huge capital challenge. Robots are expensive to build, and only the richest companies can carry large -scale experiences. Even the most efficient human robots now cost tens of thousands of dollars, which makes group adoption simply unrealistic.

Pubble Grade 5: Evaluation Effectiveness

This is the “invisible” bottle. Think about it, you can test great artificial intelligence models online such as Chatgpt almost their functions – after issuing a new language model, can researchers or ordinary users around the world extract conclusions about their performance within a few hours. But the evaluation of material artificial intelligence requires the spread of the real world, which takes time.

Tesla Full Driving Program (FSD) is a good example. If you record a million miles without an accident, does this mean that he truly achieved the independence of the 5 level? What about 10 million miles? The problem of automatic intelligence is that the only way to verify healthy is to know where it fails in the end, which means spreading a large -scale neighborhood in the long run.

Pubble neck 6: Human Resources

The challenge that was underestimated is that human action is indispensable in the development of automatic artificial intelligence. Amnesty International alone is not enough. Robots need human operators to provide training data; Maintenance teams to maintain the operation of robots; The main researchers/developers to improve artificial intelligence models constantly. Unlike artificial intelligence models that can be trained in the cloud, robots require continuous human intervention – a major challenge that DePin should treat.

Future: When will the moment of ChatGPT come to robots?

Some believe that the moment of ChatGPT of the robots is coming. Michael is somewhat skeptical. Given the challenges of devices, data and evaluation, it is believed that the AI’s general robots are still far from collective adoption. However, the progress of DePin Robotics gives people hope. The development of robots should be decentralized, and not controlled by a few large companies. The scale and coordination of the Central Central Capital Network can be spread. Instead of relying on a large company to pay for the construction of thousands of robots, it is better to put individuals who can contribute to a joint network.

For example – first, DePin speeds up data collection and evaluates it. Instead of waiting for a company to publish a limited number of robots to collect data, decentralized networks can work in parallel and collect data on a large scale. For example, in the AI ​​competition against the human robot in Abu Dhabi, researchers from institutions such as DeepMind and UT Austin have tested artificial intelligence models against human players. Although humans still possess the upper hand, researchers were enthusiastic about the unique data collection collected from robot reactions in the real world. This indirectly proves the need for sub -networks that connect different components of robots. The enthusiasm of the research community also explains that even if the full self -rule is still a long -term goal, DePin Robotics has shown a tangible value of data collection and training to publishing and verifying the real world.

On the other hand, the AI’s devices design improvements, such as optimizing chips and material engineering with artificial intelligence, may greatly shorten the schedule. A specific example is that Frodobot Lab has worked with other institutions to ensure two NVIDIA H100 graphics processing units, each containing eight H100 chips. This provides researchers with the computing power needed to address artificial intelligence models and improve the real data collected from robot spread. Without these computing resources, the most valuable data sets cannot be used. It can be seen that by accessing the DePin’s decentralized computing infrastructure, robotics can enable researchers all over the world to train and evaluate models without restricting thick capital GPU ownership. If DePin can successfully advance data and apply to devices, the future of robots may come sooner than expected.

In addition, artificial intelligence agents such as SAM, which is the KOL TREVEL with meme, displays new profit models for decentralized robots. SAM works independently, over 24/7 in multiple cities while the distinctive MEME symbols are estimated at the value. This model shows how the smart robots that DePin works can maintain themselves financially through decentralized ownership and symbolic incentives. In the future, artificial intelligence agents can use these symbols to push human operators to help, hire additional automatic assets, or bids in tasks in the real world, creating an economic cycle that benefits from the development of artificial intelligence and the participants.

The final summary

The development of Robot Ai depends not only on algorithms, but also on devices promotions, data accumulation, financial support and human participation. In the past, the development of robotics industry was limited at high costs and the domination of large companies, which hindered the pace of innovation. The creation of the DePin Robot network means that with the power of decentralized networks, robot data collection, computing resources, and capital investment can be coordinated on a global scale, which not only speeds up artificial intelligence training and improvement of devices, but also reduces the development threshold and allows more researchers, businessmen and individual users. We also hope that the robotics industry will no longer rely on a few technology giants, but will be jointly driven by the global community to move towards a really open and sustainable ecosystem.