Why is the adoption of AI in the automotive industry stalling?

In the automotive industry today, AI There is a growing need to utilize this.
Grand View Research According to a survey by AI The market 2030 By year 1,472 billion US dollars (approx. 21.7 The market is expected to grow to a trillion yen (trillion yen) scale, and investments are underway to make vehicles more intelligent and improve comfort (*1).

*1 Source: Grand View Research, Automotive Artificial Intelligence Market Size, Share & Trends Report, 2023–2030

Automotive Artificial Intelligence Market | Industry Report 2030

However, in actual practice, the following challenges are hindering the introduction of AI:

・It works in PoC (proof of concept), but cannot be moved to mass production.

・Even though cloud-based AI is highly efficient, it is not suitable for in-vehicle units due to limitations on power consumption and heat dissipation.

・Due to a shortage of AI personnel, it is difficult to transition from prototyping to mass production

As a result, "AI implementation ending at PoC" has become a common concern across the industry.

AI is not just about the cloud; it's also about using AI on-site

Traditionally, AI has mainly focused on learning and inference in a cloud environment.

Meanwhile, a new option known as "physical AI" has been gaining attention in recent years.

Physical AI is a technology that recognizes and understands the real world through sensors and actuators, and physically executes the results of AI decisions.

So, what are the differences between physical AI and traditional cloud-based AI? The table below summarizes the differences.

By using physical AI and cloud-based AI depending on the application and environment, it is possible to make the most of the strengths of each.

Physical AI is small and easy to incorporate into the workplace, and because processing is done locally, it has low latency and excellent security. There is little difference in the environment from PoC to mass production, making implementation smooth.

While cloud-based AI can utilize large-scale computing resources, it relies on communications, which can lead to delays and risks of information leaks, and may require restructuring when deployed in production.

Automotive application areas - starting with peripheral use cases

In core areas of autonomous driving and ADAS (lane keeping assist, collision avoidance braking, etc.), AI is already being implemented using GPUs and dedicated SoCs. However, these technologies are subject to high safety requirements that are directly linked to functional safety, making it difficult to immediately adopt emerging technologies.

On the other hand, in peripheral use cases that are not directly related to functional safety, AI can be introduced more flexibly, and it is an area where the benefits of physical AI can be easily utilized.

<Example>

- In-vehicle monitoring (detecting drowsiness, posture and number of occupants)

Parking assistance and perimeter monitoring (blind spot completion, moving object detection)

Entertainment and comfort control (eye tracking, gesture detection, voice assistant)

These areas allow for quick delivery and low cost, making it easy to launch PoCs and smoothly expand to mass production. By first introducing AI in peripheral areas, you can practically and steadily expand the use of AI.

The features of physical AI and SiMa.ai, the platform that makes it possible

Physical AI has the advantage of being easy to use in the field, something that conventional cloud-based AI does not have.

• Size: Mainly small modules, allowing direct integration into equipment or on-site
• Real-time performance: Local processing makes it easy to realize decisions and control with low latency
• Ease of implementation: The same environment can be used from development to mass production, allowing for a smooth transition

​SiMa.ai is an actual product that embodies these features.

The company's platform has the following features:

1. High-efficiency AI chip: Achieves approximately 50 trillion calculations per second with power consumption equivalent to 5W
2. High real-time performance and security: No need for cloud communication, resulting in low latency and reduced risk of information leaks
3. Development environment: GUI tools allow drag-and-drop operation, even for non-experts
4. Integrated operation: Providing a comprehensive platform of optimized hardware and software

​SiMa.ai offers practical solutions to the challenges faced when implementing AI in manufacturing.

Summary: "Physical AI" is a new option for in-vehicle AI

While the adoption of in-vehicle AI is progressing, cloud-based AI has faced challenges such as power consumption and the transition to mass production, resulting in limited proof-of-concept (PoC) capabilities. This is where physical AI, which is centered around small modules, offers low latency, and allows for smooth implementation, is gaining attention.

In particular, in peripheral areas such as in-vehicle monitoring and parking assistance, the process from PoC to mass production can be smoothly implemented, and the barriers to implementation are low.

SiMa.ai 's highly efficient AI chips and GUI development environment make on-site AI a reality. As your first step in introducing AI to your vehicle, why not consider the new option of physical AI?

Inquiry

Please feel free to contact us with any questions about our products, technical inquiries, sample requests, or estimates.