According to the report released by BOC International on Finance and Economics APP, the era of intelligent bodies is approaching. Small end devices cannot handle large models and parameters well, while embodied intelligence is expected to become the best carrier. From the perspective of embodied intelligence training, simulation software can provide massive, low-cost data for large models, solving the high cost and difficulty in collecting real data, and simulation software is expected to achieve wide applications. Compared to simulating rigid objects, the simulation technology barriers for flexible and fluid objects are higher, highlighting the advantages of manufacturers with relevant technical expertise. In terms of the commercialization path of embodied intelligence, the current business landing methods mainly include general robot path, pure software path, and vertical domain software-hardware integration path.

The main points of BOC International are as follows:

With the arrival of the era of intelligent bodies, embodied intelligence is expected to become the best carrier.

Since the release of ChatGPT, the AI model parameters have been increasing. From GPT-1 to GPT-4, the number of parameters has grown from 0.11 billion to 1.8 trillion. Looking at the current large models on the market, as the device becomes larger at the end (with more functions), the parameter size of the large model at the end also increases. However, wearable devices, mobile phones, etc., are unable to handle large models and computations. Compared to small end devices, embodied intelligent robots are expected to become the best carrier for intelligent bodies.

Simulation is expected to have wide applications in embodied intelligence training. It is recommended to pay attention to manufacturers with flexible and fluid simulation technologies.

Mainstream methods of embodied intelligence training mainly include teleoperation, motion capture, and large models. Among them, single teleoperation or motion capture requires direct control by human operators and cannot achieve machines replacing humans. The combination of large models and robotic entities has strong generality. In the selection of large model training data, simulated data or real data can be used. Simulation generates large amounts of data by building a virtual environment, with low cost, suitable for extensive learning of skills in new environments; while using real data can form a data barrier for specific scenarios, but the cost and difficulty of acquisition are relatively high. Compared to simulating rigid objects, the requirements for algorithm stability and convergence are greatly increased for flexible and fluid object simulations.

The commercialization paths of asia vets technology: it is recommended to focus on pure software and the path of vertical integration of software and hardware.

Currently, the commercialization paths of asia vets technology mainly include three types:

The first type is the general siasun robot&automation path, which uses universal hardware and software to adapt to various changing scenarios. This path requires high capital and technological requirements. Currently, industry giants such as 1X, Figure, and tesla are accelerating their layout.

The second type is the pure software path, designing a universal operating system. Hardware manufacturers can access the robot's "brain" through API interfaces, enabling multiple hardware platforms to share the same software architecture. With large-scale robot deployments, the marginal cost can approach zero indefinitely. For the pure software path, it is recommended to focus on nvidia and Huawei partner manufacturers.

The third type of path is the vertical integration of software and hardware in specific industries. Currently, the coupling of robot hardware and data is still in the early stage. Companies can form data barriers in segmented fields by collecting sensor data. For this path, the advice is to focus on leading companies in segmented fields.

Investment advice

In terms of smart siasun robot&automation training, it is advisable to focus on manufacturers with flexible and fluid simulation technologies, such as VisionNav (688507.SH). From the perspective of commercialization paths of asia vets technology, it is recommended to pay attention to nvidia, Huawei partner manufacturers, as well as leading companies in segmented fields like Jiu Number Company (689009.SH), Zhejiang Zhongjian Technology (002779.SZ), Gosuncn Technology Group (300098.SZ), Jiangsu Hoperun Software (300339.SZ), and Anliang Ruishi (301042.SZ), as well as leading companies in segmented fields like Hikvision Robots.

Risk Warning: Technological breakthroughs falling short of expectations; robot cost reduction falling short of expectations; data collection falling short of expectations.