On February 27, Messari hosted a podcast on "Building Decentralized Physical Artificial Intelligence," inviting Michael Cho, co-founder of FrodoBot Lab. They discussed the challenges and opportunities of Decentralized Physical Infrastructure Networks (DePIN) in the field of robotics. Although this field is still in its infancy, it has enormous potential and could fundamentally change how AI robots operate in the real world. However, unlike traditional AI that relies heavily on vast amounts of internet data, DePIN robotic AI technology faces more complex issues, such as data collection, hardware limitations, evaluation bottlenecks, and the sustainability of economic models.
In today's article, we will break down the key points from this discussion, examine the problems faced by DePIN robotic technology, identify the main obstacles to expanding decentralized robotics, and explore why DePIN has advantages over centralized approaches. Finally, we will discuss the future of DePIN robotic technology and see if we are on the verge of a "ChatGPT moment" for DePIN robotics.
Where are the bottlenecks in DePIN intelligent robots?
When Michael Cho first started FrodoBot, the biggest headache was the cost of robotic technology. Commercial robots on the market are prohibitively expensive, making it difficult to promote AI applications in the real world. His initial solution was to create a low-cost autonomous robot priced at just $500, intending to win with a price advantage over most existing projects.
However, as he and his team delved deeper into research and development, Michael realized that cost was not the real bottleneck. The challenges of Decentralized Physical Infrastructure Networks (DePIN) in robotics are far more complex than simply being "expensive." As FrodoBotLab progressed, several bottlenecks in DePIN robotic technology gradually came to light. To achieve large-scale deployment, the following bottlenecks must be overcome.
Bottleneck One: Data
Unlike 'online' AI models trained on vast amounts of internet data, embodied AI needs to interact with the real world to develop intelligence. The problem is that there is currently no large-scale infrastructure for this, and there is no consensus on how to collect this data. Data collection for embodied AI can be categorized into three main types:
▎The first type is human operation data, which is generated when humans manually control robots. This type of data is of high quality and can capture video streams and action labels—essentially what humans see and how they respond. This is the most effective way to train AI to mimic human behavior, but the downside is that it is costly and labor-intensive.
▎The second type is synthetic data (simulated data), which is useful for training robots to move in complex terrains, such as walking on rugged ground, and is particularly useful in specialized fields. However, for tasks that are highly variable, such as cooking, simulated environments fall short. We can imagine the scenario of training a robot to fry an egg: slight variations in the type of pan, oil temperature, and room conditions can affect the outcome, and virtual environments struggle to cover all scenarios.
▎The third type is video learning, where AI models learn by observing videos of the real world. Although this method has potential, it lacks the direct interactive feedback that true intelligence requires.
Bottleneck Two: Level of Autonomy
Michael mentioned that when he first tested FrodoBot in the real world, it was primarily used for last-mile delivery. From the data, the results were actually quite good—the robot successfully completed 90% of delivery tasks. However, a 10% failure rate in real life is unacceptable. A robot that fails once in every ten deliveries cannot be commercialized. Just like automated driving technology, a self-driving car can have a record of ten thousand successful drives, but one failure is enough to undermine consumer confidence.
Therefore, for robotic technology to be truly practical, the success rate must approach 99.99% or even higher. The problem is that every 0.001% increase in accuracy requires exponentially more time and effort. Many people underestimate the difficulty of this final step.
Michael recalled that in 2015, when he sat in a prototype of Google's self-driving car, he felt that fully autonomous driving was just around the corner. A decade later, we are still discussing when we can achieve Level 5 autonomy. The progress of robotic technology is not linear but exponential—each step forward significantly increases the difficulty. Achieving that final 1% accuracy may take years or even decades.
Bottleneck Three: Hardware: AI Alone Cannot Solve the Robotics Problem
Even if AI models are powerful, existing robotic hardware is not yet ready to achieve true autonomy. For example, a commonly overlooked hardware issue is the lack of tactile sensors—current best technologies, such as Meta AI's research, still fall far short of the sensitivity of human fingertips. Humans interact with the world through vision and touch, while robots know almost nothing about texture, grip, and pressure feedback.
There is also the occlusion problem—when an object is partially blocked, it is difficult for robots to recognize and interact with it. Humans can intuitively understand an object even if they cannot see its entirety.
In addition to perception issues, robotic actuators themselves have flaws. Most humanoid robots place actuators directly at the joints, making them bulky and potentially dangerous. In contrast, the human tendon structure allows for smoother and safer movements. This is why existing humanoid robots appear stiff and inflexible. Companies like Apptronik are developing more biologically inspired actuator designs, but these innovations will take time to mature.
Bottleneck Four: Why is Hardware Scaling So Difficult?
Unlike traditional AI models that rely solely on computational power, the implementation of intelligent robotic technology requires the deployment of physical devices in the real world. This presents significant capital challenges. Building robots is expensive, and only the wealthiest large companies can afford large-scale experiments. Even the most efficient humanoid robots currently cost tens of thousands of dollars, making large-scale adoption unrealistic.
Bottleneck Five: Evaluating Effectiveness
This is an "invisible" bottleneck. Consider that online AI models like ChatGPT can almost instantaneously test their functionality—once a new language model is released, researchers or ordinary users worldwide can generally conclude its performance within hours. However, evaluating physical AI requires real-world deployment, which takes time.
Tesla's Full Self-Driving (FSD) software is a good example. If Tesla records one million miles without an accident, does that mean it has truly achieved Level 5 autonomy? What about ten million miles? The problem with robotic intelligence technology is that the only way to validate it is to see where it ultimately fails, which means large-scale, long-term real-time deployment.
Bottleneck Six: Human Labor
Another underestimated challenge is that human labor remains indispensable in the development of robotic AI. AI alone is not enough. Robots need human operators to provide training data; maintenance teams to keep the robots running; and essential researchers/developers to continuously optimize AI models. Unlike AI models that can be trained in the cloud, robots require ongoing human intervention—this is also a major challenge that DePIN must address.
The Future: When Will the ChatGPT Moment for Robotics Arrive?
Some believe that the ChatGPT moment for robotics is imminent. Michael is somewhat skeptical about this. Given the challenges of hardware, data, and evaluation, he believes that general robotic AI is still far from large-scale adoption. However, the progress of DePIN robotic technology indeed offers hope. The development of robotic technology should be decentralized rather than controlled by a few large companies. The scale and coordination of a decentralized network can distribute the capital burden. Instead of relying on a single large company to fund the construction of thousands of robots, it may be better to include individuals who can contribute into a shared network.
For example—first, DePIN accelerates data collection and evaluation. Instead of waiting for a company to deploy a limited number of robots to collect data, a decentralized network can operate in parallel on a larger scale to gather data. For instance, in a recent AI and human-robot competition in Abu Dhabi, researchers from institutions like DeepMind and UT Austin tested their AI models against human players. While humans still had the upper hand, researchers were excited about the unique dataset collected from real-world robot interactions. This indirectly demonstrates the need for subnetworks that connect various components of robotic technology. The enthusiasm in the research community also indicates that even if full autonomy remains a long-term goal, DePIN robotic technology has already shown tangible value from data collection and training to real-world deployment and validation.
On the other hand, AI-driven hardware design improvements, such as optimizing chips and materials engineering with AI, could significantly shorten timelines. A specific example is FrodoBot Lab's collaboration with other institutions to secure two boxes of NVIDIA H100 GPUs—each box containing eight H100 chips. This provides researchers with the necessary computational power to process and optimize AI models based on real-world data collected from robot deployments. Without such computational resources, even the most valuable datasets cannot be fully utilized. It is evident that access to decentralized computing infrastructure through DePIN allows robotic technology networks to enable researchers worldwide to train and evaluate models without being constrained by capital-intensive GPU ownership. If DePIN can successfully crowdsource data and hardware advancements, the future of robotic technology may arrive sooner than expected.
Additionally, AI agents like Sam (a travel KOL robot with meme coins) demonstrate a new profit model for decentralized robotic technology networks. Sam operates autonomously, live-streaming 24/7 in multiple cities, while its meme coins appreciate in value. This model showcases how DePIN-driven intelligent robots can sustain their finances through decentralized ownership and token incentives. In the future, these AI agents could even use tokens to pay human operators for assistance, rent additional robotic assets, or bid for real-world tasks, creating an economic cycle that benefits both AI development and DePIN participants.
Conclusion
The development of robotic AI depends not only on algorithms but also on hardware upgrades, data accumulation, funding support, and human participation. In the past, the development of the robotics industry was limited by high costs and the dominance of large enterprises, hindering the pace of innovation. The establishment of DePIN robotic networks means that, with the power of decentralized networks, robotic data collection, computational resources, and capital investment can be coordinated globally, accelerating AI training and hardware optimization while lowering development barriers, allowing more researchers, entrepreneurs, and individual users to participate. We also look forward to a robotics industry that no longer relies on a few tech giants but is driven by a global community towards a truly open and sustainable technological ecosystem.
All content on the Coinspire platform is for reference only and does not constitute an offer or advice for any investment strategy. Any personal decisions made based on the content of this article are the sole responsibility of the investor, and Coinspire is not responsible for any profits or losses arising therefrom.
Investment involves risks; decisions should be made cautiously.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
