Pantera Partner: The Era of AI Bots Driven by Encryption

Author: Paul Veradittakit, Partner at Pantera Capital; Translated by: Jinse Caijing xiaozou

Abstract:

VLA* Innovation and economies of scale are driving the emergence of affordable, efficient, and versatile humanoid robots.*

As warehouse robots expand into the consumer robot market, the safety, financing, and evaluation mechanisms of robots are worth exploring in depth.

Cryptography will drive the development of the robotics industry by providing economic guarantees for the security of robots and optimizing their docking infrastructure, latency, and data collection processes.

ChatGPT has completely rewritten humanity's expectations of artificial intelligence. When large language models began to interact with the external software world, many believed that AI agents represented the ultimate form. However, if we look back at classic sci-fi films like "Star Wars," "Blade Runner," or "RoboCop," we can see that what humanity truly dreams of is artificial intelligence being able to interact with the physical world in the form of robots.

According to Pantera Capital, the "ChatGPT moment" in the robotics field is approaching. We will first analyze how breakthroughs in artificial intelligence over the past few years have changed the industry landscape, and then discuss how battery technology, latency optimization, and data collection improvements will shape the future picture, as well as the role of cryptographic technology in it. Finally, we will explain why we believe that robot safety, financing, valuation, and education are verticals that require focused attention.

1**, Factors of Change**

(1) Breakthrough in Artificial Intelligence

Advancements in the field of multimodal large language models are providing robots with the "brain" needed to perform complex tasks. Robots primarily perceive their environment through two senses: vision and hearing.

Traditional computer vision models (such as convolutional neural networks) excel at object detection or classification tasks but struggle to translate visual information into purposeful action commands. Large language models perform excellently in text understanding and generation but are limited in their perception of the physical world.

Through the Visual-Language-Action model (VLA), robots can integrate visual perception, language understanding, and physical actions within a unified computational framework. In February 2025, Figure AI released the Helix, a universal humanoid robot control model. This VLA model sets a new benchmark for the industry with its zero-shot generalization capability and the System 1/System 2 dual architecture. The zero-shot generalization feature allows robots to instantly adapt to new scenarios, objects, and commands without the need for repetitive training for each task. The System 1/System 2 architecture separates high-level reasoning from lightweight reasoning, enabling commercial humanoid robots to achieve both human-like thinking and real-time precision.

(2) Economical robots become a reality**

The technology that changes the world has a common characteristic - accessibility. Smartphones, personal computers, and 3D printing technology have all become widespread at prices affordable to the middle class. When robots like the Unitree G1 are priced lower than a Honda Accord or the minimum annual income of $34,000 in the U.S., it is not surprising to imagine a world where physical labor and daily tasks are primarily accomplished by robots.

(3) From Warehousing to Consumer-Level Market

Robotics technology is expanding from warehouse solutions to the consumer sector. This world is designed for humans—humans can perform all the tasks of specialized robots, while specialized robots cannot perform all tasks of humans. Robotics companies are no longer limited to manufacturing robots for factories, but are instead developing more versatile humanoid robots. Therefore, the forefront of robotics technology is not only in warehouses but will also penetrate everyday life.

Cost is one of the main bottlenecks of scalability. The metric we focus on the most is the comprehensive cost per hour, calculated as the sum of the opportunity cost of training and charging time, task execution costs, and the cost of purchasing robots, divided by the total operating hours of the robots. This cost must be below the average wage level of the relevant industry to be competitive.

To fully penetrate the warehousing sector, the comprehensive cost of robots must be below $31.39 per hour. In the largest consumer market—the private education and health services sector, this cost must be controlled to below $35.18. Currently, robots are developing towards being cheaper, more efficient, and more versatile.

2**, The Next Breakthrough in Robotics Technology**

(1) Battery Optimization

Battery technology has always been a bottleneck for user-friendly robots. Early electric vehicles like the BMW i3 struggled to gain popularity due to limitations in battery technology, which resulted in short range, high costs, and low practicality. Robots are facing the same predicament. Boston Dynamics' Spot robot has a single charge lasting only 90 minutes, while the Unitree G1 has a battery life of about 2 hours**. Users are clearly unwilling to manually charge every two hours**, making autonomous charging and docking infrastructure a key development direction. Currently, there are primarily two modes for charging robots: battery replacement or direct charging.

The battery replacement mode achieves continuous operation by quickly replacing the depleted battery pack, minimizing downtime, and is suitable for outdoor or factory scenarios. This process can be performed manually or automated.

Inductive charging uses a wireless power supply method. Although it takes a longer time to fully charge, it can easily achieve a fully automated process.

(2) Delay Optimization

Low-latency operations can be divided into two categories: environmental awareness and remote control. Awareness refers to the robot's spatial cognition ability of the environment, while remote control specifically refers to the real-time control by human operators.

According to research by Cintrini, robotic perception systems start with inexpensive sensors, but the technological moat lies in the integration of software, low-power computing, and millisecond-level precision control loops. Once the robot completes spatial localization, lightweight neural networks will label elements such as obstacles, pallets, or humans. After the scene labels are input into the planning system, motor commands are immediately generated and sent to the feet, wheel groups, or robotic arms. A perception delay of less than 50 milliseconds is equivalent to human reflex speed — any delay beyond this threshold will result in clumsy robotic movements. Therefore, 90% of decisions need to be made locally through a single vision-language-action network.

Fully autonomous robots must ensure that the latency of the high-performance VLA model is below 50 milliseconds; for remotely controlled robots, the signal latency between the operation end and the robot must not exceed 50 milliseconds. The importance of the VLA model is particularly highlighted here—if visual and textual inputs are processed by different models before being input into a large language model, the overall latency will far exceed the 50-millisecond threshold.

(3) Data Collection Optimization

There are mainly three ways of data collection: real-world video data, synthetic data, and remote control data. The core bottleneck between real data and synthetic data lies in bridging the gap between the physical behavior of robots and video / simulation models. Real video data lacks physical details such as force feedback, joint motion errors, and material deformation; synthetic data, on the other hand, lacks unpredictable variables such as sensor failures and friction coefficients.

The most promising data collection method is remote control—where human operators remotely control robots to perform tasks. However, labor costs are the main constraint on remote-controlled data collection.

Customized hardware development is also providing new solutions for high-quality data collection. Mecka combines mainstream methods with customized hardware to collect multidimensional human movement data, which is processed and converted into datasets suitable for training robotic neural networks, providing massive high-quality data for AI robot training with a rapid iteration cycle. These technical pipelines together shorten the conversion path from raw data to deployable robots.

3**, Key Exploration Areas**

(1) Integration of Cryptography and Robotics

Cryptographic technology can incentivize trustless parties to enhance the efficiency of robotic networks. Based on the key areas mentioned above, we believe that cryptographic technology can improve efficiency in three aspects: integration of infrastructure, latency optimization, and data collection.

Decentralized Physical Infrastructure Network (DePIN) is expected to revolutionize charging infrastructure. When humanoid robots operate globally like cars, charging stations need to be as accessible as gas stations. Centralized networks require huge upfront investments, while DePIN distributes costs among node operators, allowing charging facilities to rapidly expand into more areas.

DePIN can also utilize distributed infrastructure to optimize remote control latency. By aggregating geographically dispersed edge node computing resources, remote control commands can be processed by local or nearest available nodes, minimizing data transmission distance and significantly reducing communication latency. However, current DePIN projects mainly focus on decentralized storage, content distribution, and bandwidth sharing. Although some projects showcase the advantages of edge computing in streaming media or the Internet of Things, it has yet to extend to robotics or remote control fields.

Remote control is the most promising data collection method, but the cost of centralized entities hiring professionals to collect data is extremely high. DePIN addresses this issue by incentivizing third parties with crypto tokens to provide remote control data. The Reborn project builds a global network of remote operators, transforming their contributions into tokenized digital assets, creating a permissionless decentralized system—participants can earn rewards while also participating in governance and contributing to AGI robot training.

(2) Security has always been a core concern

The ultimate goal of robotics is to achieve full autonomy, but as the "Terminator" series of films warns, humanity is most reluctant to see autonomy turn robots into offensive weapons. The safety issues of large language models have raised concerns, and when these models have physical action capabilities, robot safety becomes a key prerequisite for societal acceptance.

Economic security is one of the pillars of a thriving robotic ecosystem. OpenMind, a company in this field, is building FABRIC—a decentralized machine coordination layer that achieves device identity authentication, physical presence verification, and resource acquisition through cryptographic proofs. Unlike simple task market management, FABRIC enables robots to autonomously prove their identity information, geographical location, and behavior records without relying on centralized intermediaries.

Behavioral constraints and identity authentication are executed through on-chain mechanisms, ensuring that anyone can audit compliance. Robots that meet safety standards, quality requirements, and regional regulations will be rewarded, while violators will face penalties or disqualification, thus establishing an accountability and trust mechanism within the autonomous machine network.

Third-party re-staking networks (such as Symbiotic) can also provide equivalent security guarantees. Although the penalty parameter system still needs to be improved, the relevant technology has entered a practical stage. We expect industry security guidelines to soon take shape, at which point the penalty parameters will be modeled based on these guidelines.

Implementation Plan Example:

Robot company joins the Symbiotic network.

Set verifiable forfeiture parameters (e.g. "apply human contact force exceeding 2500 Newtons");

Stakers provide margin to ensure the robot adheres to parameters;

In the event of a violation, the collateral will be used as compensation for the victim.

This model incentivizes companies to prioritize security while promoting consumer acceptance through the insurance mechanism of the staked fund pool.

The Symbiotic team's insights into the field of robotics are:

Symbiotic* The Universal Staking Framework aims to extend the concept of staking to all areas that require economic security endorsement, whether through shared or independent models. Its application scenarios need specific case designs from insurance to robotics. For example, a robotic network can be fully built based on the Symbiotic framework, enabling stakeholders to provide economic guarantees for network integrity.*

4**, Filling the Gaps in the Robotics Technology Stack**

OpenAI has promoted the popularization of AI, but the cornerstone of ChatGPT has long been established. Cloud services have broken the model's dependence on local computing power, Huggingface has achieved model open-sourcing, and Kaggle provides an experimental platform for AI engineers. These incremental breakthroughs have collectively contributed to the democratization of AI.

**Unlike AI, the robotics field is difficult to enter with limited funding. To achieve widespread adoption of robotics, the development threshold must be reduced to a level of convenience similar to that of AI application development. We believe there is room for improvement in three areas: financing mechanisms, evaluation systems, and educational ecosystems.

Funding is a pain point in the field of robotics. Developing a computer program only requires a computer and cloud computing resources, while building a fully functional robot necessitates the procurement of hardware such as motors, sensors, and batteries, with costs easily exceeding $100,000. This hardware nature makes robot development less flexible and significantly more expensive compared to AI.

The evaluation infrastructure for robots in real-world scenarios is still in its infancy. The AI field has established a clear loss function system, and testing can be fully virtualized. However, excellent virtual strategies cannot be directly translated into effective solutions in the real world. Robots need to test the evaluation facilities of autonomous strategies in diverse real-world environments to achieve iterative optimization.

As these infrastructures mature, talent will flood in, and humanoid robots will replicate the explosive curve of Web2. The crypto robotics company OpenMind is advancing in this direction—its open-source project OM1 ("robotic version of the Android system") transforms raw hardware into economically aware, upgradeable agents. Visual, language, and motion planning modules can be plug-and-play like mobile applications, with all reasoning steps presented in clear English, allowing operators to audit or adjust behaviors without touching the firmware. This natural language reasoning capability enables a new generation of talent to seamlessly enter the robotics field, making a critical step towards igniting an open platform for the robotics revolution, much like the accelerating effect of the open-source movement on AI.

Talent density determines industry trajectory. A structured inclusive education system is crucial for talent delivery in the robotics field. The listing of OpenMind on Nasdaq marks the beginning of a new era where intelligent machines simultaneously participate in financial innovation and physical education. OpenMind and Robostore jointly announced the launch of the first general education curriculum based on the Unitree G1 humanoid robot in public K-12 schools in the United States. **The curriculum is designed to be platform-agnostic, adaptable to various robotic forms, and provides students with practical operational opportunities. This positive signal reinforces our judgment: **In the coming years, the richness of educational resources in robotics will be on par with the AI field. **

5**, Future Outlook**

The innovations and economies of scale of the Visual-Language-Action model (VLA) have given rise to affordable, efficient, and versatile humanoid robots. As warehouse robots expand into the consumer market, safety, financing models, and evaluation systems have become key areas of exploration. We firmly believe that cryptographic technology will drive the development of robots through three pathways: providing economic guarantees for safety, optimizing charging infrastructure, and enhancing latency performance and data collection pipelines.