For many companies, working on AI models to be used in the physical world, it is data and data and more data as the largest opportunity. As nicely labeled and clean real world data are as readily available as hen's teeth. The cost and effort spent gathering and cleaning up that data can be enormous as well.
Bifrost, a 3D data-generation platform, thinks its technology can help robotics and industrial companies solve at least one part of that problem: the time required to train AI models. Based out of San Francisco, the startup says its platform lets companies generate simulated 3D worlds to instruct their AI models and help their robots adjust to new objects, tasks, and surroundings within hours instead of months.
The company, on Wednesday, disclosed it had raised $8 million in a Series A round of funding led by Carbide Ventures.
"Most of our customers need vast amounts of real-world data to train AI models," said Wong, co-founder and chief executive officer, speaking to TechCrunch in an interview. This often means they would have to deploy fleets of robots across hundreds of locations, collect millions of hours of footage, manually label the data, and implement rigorous quality checks to reduce human errors and bias. This approach is brutal. It costs millions, takes years, and proves nearly impossible to scale.
Wong co-founded the company with Aravind Kandiah in 2020. Prior to this, Wong was working on AI perception models for self-driving cars at NuTonomy, the MIT spinout working on self-driving vehicles and mobile robots. Kandiah had previously built a medical AI system that detects early signs of blindness and diabetic retinopathy.
"It didn't take us long to realize the following: AI and robotics need enormous amounts of high-quality data to function well, and that data is so critical," Kandiah told TechCrunch. "It can make or break the performance and potential of these systems. So, we joined forces to form Bifrost with just one goal in mind - to solve the data problem so AI and robotics may finally tackle the complex issues in the physical world.
Bifrost claims to be different from its competitors in that its platform does not require a team of 3D simulation experts to create such data. This, Wong said, gives AI engineers a huge advantage in terms of being able to develop AI systems for tasks such as patrolling contested waters with autonomous boats without needing to hire a 3D team.
"Nvidia's Omniverse tools, by contrast, require a dedicated 3D team just to operate," Kandiah said, adding that Bifrost enables engineers in various heavy industries to teach AI systems new skills and accomplish more faster.
Bifrost's product is currently closed beta with select heavy industry partners. The company plans to use the fresh capital to fund the public launch of the platform in the near future and hire more employees to speed up product development.
According to Wong, the target market for this company is the U.S. However, it also starts gaining speed in Japan with such a large industrial sector in the country. A revenue model of this start-up is through an annual subscription model.
AI development will be the main customer segment, as they specialize in robotics systems and other similar computer vision and perception models and will be used across industry verticals such as aerospace and defense, maritime and geospatial, robotics and industrial automation. These users are large industrial houses and government organizations, though will also include startups during their growth to late stage as these will have the respective teams working on AI/ML-based physical product innovations in their domain space.
"We're first focused on mission-critical, heavy industrial applications. By 2025, we expect to see increased availability of the platform. Going forward, we intend to support a broad range of commercial robotics use cases, especially as applications in robotics have been proliferating rapidly across almost every major sector and industry," Kandiah said.
This funding round takes the total capital raised for Bifrost to $13.7 million. The round saw participation from Airbus Ventures, Peak XV's Surge, Wavemaker Partners, MD One, and Techstars. The outfit has 22 staff in the U.S. and Singapore.