The integration of DePIN and embodied intelligence: technical challenges and future prospects

Introduction: On February 27, Messari held a podcast on “Building Decentralized Physical Artificial Intelligence”, inviting Michael Cho, co-founder of FrodoBot Lab. They talked about the challenges and opportunities of the Decentralized Physical Infrastructure Network (DePIN) in the field of robotics. Although this field is still in its infancy, it has great potential and may revolutionize the way AI robots operate in the real world. However, unlike traditional AIs that rely on large amounts of Internet data, the problems faced by DePIN robotics AI technology are more complex, such as data collection, hardware restrictions, evaluation bottlenecks, and the sustainability of economic models.

In today's article, we will break down the key points in this discussion to see what problems DePIN robotics encounters, what are the main obstacles to expanding decentralized robots, and why DePIN has more advantages than centralized methods. Finally, we will also discuss the future of DePIN robotics to see if we are about to usher in the "ChatGPT moment" of DePIN robotics.

What is the bottleneck of DePIN intelligent robots?

When Michael Cho first started working on FrodoBot, the most troublesome thing was the cost of robotics. The prices of commercial robots on the market are ridiculously high, making it difficult to promote AI application products in the real world. His initial solution was to build a low-priced autonomous robot that costs only $500, and intend to win at a price advantage that is cheaper than most existing projects.

But as he and his team developed in depth, Michael realized that cost was not a real bottleneck. The challenges of decentralized physical infrastructure networks (DePINs) in robotics are far more complex than "not expensive or not". With the continuous advancement of FrodoBotLab, multiple bottlenecks in DePIN robotics have gradually surfaced. To achieve large-scale deployment, the following bottlenecks must be overcome.

Bottleneck 1: Data

Unlike the 'online' AI models trained on a large amount of Internet data, embodied AI (embodied AI) needs to interact with the real world to develop intelligence. The problem is that the world currently does not have such a large-scale foundation, and there is no consensus on how to collect this data. Data collection of embodied AI can be classified into the following three categories:

▎The first category is human operation data, which is the data generated when humans manually control robots. This type of data is of high quality and can capture video streams and action tags - that is, what humans see and how to respond accordingly. This is the most effective way to train AI to imitate human behavior, but the disadvantage is that it is high cost and labor intensity.

▎The second category is synthetic data (simulated data). This type of data is useful for training robots to move in complex terrain. For example, training robots to walk on rough ground, which is very useful for some specialized fields. But for some changing tasks, such as cooking, it is not good to simulate the environment. We can imagine the situation where a robot is trained to fry eggs: slight changes in the type of pot, oil temperature, and room conditions will affect the results, and it is difficult for the virtual environment to cover all scenarios.

▎The third category is video learning, which means letting AI models learn by observing videos in the real world. While this approach has potential, it lacks the true physics direct interactive feedback required by intelligence.

Bottleneck 2: Autonomy level

Michael mentioned that when he first tested FrodoBot in the real world, he mainly used robots to do last-mile delivery. Judging from the data, the result is actually pretty good - the robot successfully completed 90% of the delivery tasks. But a 10% failure rate in real life is unacceptable. A robot that fails every ten delivery times cannot be commercialized at all. Just like automated driving technology, unmanned driving can have a successful driving record of 10,000 times, but one failure is enough to defeat the confidence of commercial consumers.

Therefore, to make robotics truly practical, the success rate is close to 99.99% or even higher. But the problem is that every 0.001% improvement in accuracy requires exponential time and effort. Many people underestimate the difficulty of this last step.

Michael recalls that when he sat in Google’s prototype in 2015, he felt that fully autonomous driving was about to be realized. Ten years have passed, and we are still discussing when we can achieve full autonomy in Level 5. The advancement in robotics is not linear, but exponential - every step forward, the difficulty will increase significantly. This last 1% accuracy rate may take several years or even decades to achieve.

Bottleneck 3: Hardware: AI alone cannot solve the robot problem

To take a step back, no matter how powerful the AI ​​model is, the existing robot hardware is not ready to achieve true autonomy. For example, the most easily overlooked problem in hardware is the lack of haptic sensors - the best technology at present, such as Meta AI research, is far from reaching the sensitivity of human fingertips. Humans interact with the world by vision and touch, while robots know little about texture, grip and pressure feedback.

There is also the problem of occlusion - when the object is blocked, it is difficult for the robot to recognize and interact with it. Even if humans cannot see the full picture of an object, they can understand it intuitively.

In addition to perception problems, the robot actuator itself also has flaws. Most humanoid robots place actuators directly on joints, making them bulky and potentially dangerous. In contrast, human tendon structure makes movement smoother and safer. This is why existing humanoid robots look stiff and inflexible. Companies like Apptronik are developing more bioinspired actuator designs, but these innovations will take time to mature.

Bottleneck 4: Why is hardware expansion so difficult

Unlike traditional AI models that rely solely on computing power, the implementation of intelligent robotics requires the deployment of physical devices in the real world. This brings huge capital challenges. Building robots is expensive, and only the richest large companies can afford large-scale experiments. Even the most efficient humanoid robots now cost tens of thousands of dollars, and it is simply unrealistic to popularize them on a large scale.

Bottleneck 5: Assessing effectiveness

This is a "invisible" bottleneck. Think about it, online AI models like ChatGPT can test their functions almost instantly. After a new language model is released, researchers or ordinary users around the world can basically draw conclusions about its performance within a few hours. But evaluating physical AI requires real-world deployment, which takes time.

Tesla's fully autonomous driving (FSD) software is a good example. If Tesla recorded 1 million miles without accidents, does that mean it really achieved Level 5 autonomy? What about 10 million miles? The problem with robotic intelligence technology is that the only way to verify it is to see where it ultimately fails, which means large-scale, long-term real-time deployment.

Bottleneck Six: Human Resources

Another underestimated challenge is that human labor remains indispensable in robot AI development. AI alone is not enough. Robots require human operators to provide training data; maintenance teams keep the robot running; and essential researchers/developers continuously optimize AI models. Unlike AI models that can be trained in the cloud, robots require continuous human intervention—a major challenge that DePIN must solve.

The Future: When will the ChatGPT moment arrive for robotics?

Some believe that the ChatGPT moment for robotics is coming soon. Michael was skeptical about this. Given the challenges of hardware, data and evaluation, he believes that universal robot AI is still far from mass adoption. However, the progress of DePIN robotics has indeed given people hope. The development of robotics should be decentralized, not controlled by a few large companies. The scale and coordination of a decentralized network can diversify the capital burden. Rather than relying on a large company to invest in thousands of robots, it might be better to put individuals who can contribute into a shared network.

For example - First, DePIN accelerates data collection and evaluation. Without waiting for a company to deploy limited robots to collect data, the decentralized network can run and collect data in parallel at a larger scale. For example, in a recent AI-human robot competition in Abu Dhabi, researchers from institutions such as DeepMind and UT Austin tested their AI models with human players. While humans still prevail, researchers are excited by the unique data sets collected from real-world robot interactions. This indirectly demonstrates the need for subnets connecting various components of robotics technology. The enthusiasm of the research community also shows that even if full autonomy remains a long-term goal, DePIN robotics has demonstrated tangible value from data collection and training to real-world deployment and verification.

On the other hand, AI-driven hardware design improvements, such as using AI to optimize chip and material engineering, may significantly shorten the timeline. A concrete example is that FrodoBot Lab works with other agencies to ensure two boxes of NVIDIA H100GPUs – each containing eight H100 chips. This provides researchers with the necessary computing power to process and optimize AI models of real-world data collected from robot deployments. Without such computing resources, even the most valuable data sets cannot be fully utilized. It can be seen that through access to DePIN decentralized computing infrastructure, robotics networks allow researchers around the world to train and evaluate models without being restricted by capital-intensive GPU ownership. If DePIN succeeds in crowdsourcing data and hardware advancements, the future of robotics may arrive earlier than expected.

Additionally, AI agents like Sam (a travel KOL robot with meme coins) showcase a new profit model for decentralized robotics networks. Sam runs independently, broadcasts live in multiple cities 24/7, and its meme coins are also increasing. This model shows how DEPIN-powered intelligent robots can sustain their finances through decentralized ownership and token incentives. In the future, these AI agents can even use tokens to pay for assistance from human operators, rent additional robot assets, or bid for real-world tasks, thus forming an economic cycle that is both conducive to AI development and DePIN participants.

Final summary

The development of robot AI not only depends on algorithms, but also involves hardware upgrades, data accumulation, financial support and human participation. In the past, the development of the robotics industry was limited by high costs and the dominance of large enterprises, which hindered the pace of innovation. The establishment of the DePIN robot network means that with the help of the power of decentralized networks, robot data collection, computing resources and capital investment can be carried out in a coordinated manner around the world, which not only accelerates AI training and hardware optimization, but also lowers the development threshold and allows more researchers, entrepreneurs and individual users to participate. We also expect that the robot industry will no longer rely on a few technology giants, but will be promoted by the global community to move towards a truly open and sustainable technology ecosystem.