- Published on
Rivian buys startup to turn 400,000 daily clips into 3D sims
- Authors
- Name
- DeepHorus
Rivian has acquired Third Dimension, a 15-person UK startup, to convert the more than 400,000 driving sequences its fleet collects daily into detailed 3D simulation environments for training its self-driving system. The deal, reported by TechCrunch, matters because simulation is essential for validating autonomy without waiting for rare edge cases, and it supports Rivian’s ambitious robotaxi timeline with Uber.
Third Dimension was founded in 2024 by CEO Tolga Kart, a former Tesla Autopilot simulation engineer who also worked on Call of Duty and at Parallel Domain, and cofounder Piotr Sokólski, a veteran of Wayve’s neural simulator team. The startup raised a $7 million seed round led by Felicis in October 2024, originally targeting game studios, film, the military, and autonomous vehicle companies. Rivian did not disclose the purchase price.
The entire Third Dimension team will join Rivian’s autonomy and AI group under James Philbin. Most employees are based in the UK, giving Rivian a larger London footprint. Rivian pulls footage from R1S, R1T, R2, and commercial van owners who opt into camera data sharing. Philbin said the fleet returns more than 400,000 sequences every day, but the vast majority are uneventful.
Third Dimension’s generative AI tool builds 3D environments ready for game engines. For Rivian, that means rebuilding real scenes from fleet data and altering variables—a mattress in the fast lane, a cop waving traffic through a red light, a child darting between parked cars—without waiting months for the fleet to encounter them naturally. Philbin said the technology "turns real-world data into detailed 3D environments."
Simulation is central to how Rivian trains and validates autonomous driving, Philbin said. The urgency is high: Uber agreed in March to invest up to $1.25 billion in Rivian, but only $300 million comes upfront, with the rest tied to autonomy milestones starting with 10,000 driverless R2s on Uber’s network in San Francisco and Miami from 2028. Consumer timelines have already slipped once.
Rivian is not first. Waymo unveiled its Waymo World Model in February, built on Google DeepMind’s Genie 3, to generate unseen scenarios; Tesla has shown its own neural world simulator for FSD. One Electrek commenter noted the need to fuse camera data with lidar and radar. Analysts see the deal as small and targeted, buying a team that has done this exact work at Tesla and Wayve.
But simulation is a training and testing tool, not proof. A system can ace millions of generated edge cases and still fail on a real street; public confidence will depend on companies sharing real-world safety numbers. Rivian is betting that squeezing more value from every clip will help it hit 2027 eyes-off driving and 2028 Level 4 targets. Expect more consolidation of simulation talent as the AV race intensifies.
