Beyond Autonomous Driving, into the Physical AI Market: AWS and StradVision Expand Cooperation

Based on support from Amazon Web Services (AWS) cloud infrastructure, StradVision is improving the stability of its autonomous driving vision AI software. On that foundation, the company is preparing to expand into the Physical AI market, including defense, infrastructure, and robotics.

On the 8th, StradVision held a media tour at its Hwaseong StradVision Dongtan office. At the event, AWS and StradVision shared the results of their data pipeline optimization and also unveiled their future global expansion strategy.

StradVision develops AI-based object recognition software for ADAS and autonomous driving. The company conducts its business by supplying global OEMs and Tier 1 suppliers.

AWS has identified StradVision as a promising startup from the early stages. Since then, the two companies have worked together through Activate program credits, technical architecture design consulting, and support for global market expansion, while maintaining a close partnership.

Through this partnership, AWS is supporting StradVision's growth. The photo shows media tour participants posing for a commemorative photo, and the image is courtesy of AWS.

The photo includes StradVision COO Kwon Tae-san, and the image was taken by reporter Kwon Young-seok.

In his presentation, COO Kwon cited StradVision's core competitive advantage as flexible, lightweight AI software that is not tied to expensive hardware or specific chipsets.

Kwon said the company's core solution, SVNet, can run on more than 30 chipsets, including those from Texas Instruments (TI), Axera, and Samsung.

He also said SVNet is now in mass production, having been installed in more than 50 vehicle models and over 5 million vehicles worldwide.

For its ADAS expansion strategy, StradVision has adopted a camera-sensor-centered approach with stronger price competitiveness instead of expensive LiDAR, and is using that approach to expand ADAS deployment across many mass-market vehicle models.

Kwon identified software stability as the biggest challenge. Development typically takes a little over a year, but mass-produced vehicles then remain in operation for 5 to 10 years, with hundreds of thousands of vehicles to several million vehicles on the road. He said the difficulty of fully identifying in advance the exceptions that can arise during long-term, large-scale operation is the background behind the software stability challenge.

A necessary condition for pre-verifying long-term stability is large-scale exception data that cannot be secured through real-world driving alone. Accordingly, StradVision actively introduced AWS infrastructure-based synthetic data generation technology to fill this gap. The related discussion was conducted through a Q&A with StradVision COO Kwon Tae-san and Tiffany Bloomquist, head of AWS.

StradVision introduced a use case for simulation technology based on AWS GPU infrastructure to secure traffic sign data that varies by country. In the process, it ran 100 AWS GPU instances for two months. Through this effort, it generated synthetic data for more than 1,600 types of traffic signs across 32 countries.

Using this approach, StradVision also succeeded in verifying performance in a short period of time. By supplementing what was lacking from ordinary driving data with AWS-based synthetic data generation and GPU simulation, the company rapidly secured country-specific traffic sign data and carried out performance verification as well.

Another task it is pursuing separately is securing edge case data. The targets are cases that are difficult to capture in the real world or dangerous to reproduce for safety reasons. Examples include the sudden appearance of animals and abrupt cut-ins by pedestrians and vehicles.

StradVision is using actual driving video plus 3D asset synthesis and a virtual simulation environment to support stability verification of actual mass-production software. It operates a verification system by artificially introducing objects that are rare in domestic driving, such as cows and camels, and repeatedly reproducing dangerous cut-in situations involving pedestrians and vehicles.

Based on that verification system, StradVision is mass-producing and supplying its front-camera product, Front Vision. Surround Vision for automatic parking is scheduled to enter mass production in a separate region within the year, and the company plans to complete development of Multi Vision for L3 and L4 autonomous driving in the future. After completing Multi Vision development, StradVision plans to move beyond the automotive industry into Physical AI, targeting defense, infrastructure, and robotics.

Kwon Tae-san, COO of StradVision, said that AWS is a partner that has worked closely with the company from storage utilization to securing computing resources for large-scale AI software training. He explained that the company is conducting large-scale simulations with AWS cloud resources and verifying whether the software operates without problems in diverse environments around the world.

The person in the photo is Tiffany Bloomquist, head of AWS.

AWS supports startups across the full cycle, from the initial VC stage of idea development to global market expansion. In the process, AWS operates a tailored support system that includes help with technical architecture design.

A key element of that support structure is initial credit support. Another core element is linking startups with partnerships involving large enterprise companies.

Another key support element is access to AWS Marketplace. AWS Marketplace is designed to enable certified software to be sold safely and quickly to enterprise customers around the world. The marketplace also provides security and simplifies the contracting process.

For that reason, AWS Marketplace is seen as one of the key channels for startups to expand globally. For startups, it can serve as a path to quickly supply AI solutions to large enterprises without a complicated procurement process.

StradVision has completed the advancement of its internal data pipeline and is preparing to launch a software-as-a-service product in early next year, with AWS Marketplace as the go-to-market channel. The product will operate in a public cloud environment and is designed so that customers can upload data directly and generate outputs themselves.

Tiffany Bloomquist, head of AWS, said AWS is a partner that helps startups reduce infrastructure costs when validating their initial ideas and supports long-term business scaling after product-market fit is achieved. She added that StradVision is a representative case of a company that achieved cost efficiency from the early architecture-building stage and expanded its business to global markets based on the cloud.

Source: IT DAILY · Kwon Young-seok
Original: https://www.itdaily.kr/news/articleView.html?idxno=241484

References

• https://www.itdaily.kr/news/articleView.html?idxno=241484