Revv acquires Third Dimension, adding 400,000 daily scenes to build world models for training Uber vehicles

Revv made a full acquisition of the UK start-up Third Dimension, a move aimed at shifting autonomous-driving engineering from synthetic simulation environments built on video-game engines to real-world-based 3-D models. The undisclosed-value deal folds the start-up’s 15-person team, most based in the United Kingdom, into Revv’s AI and autonomous-driving group led by James Philbin, while expanding the company’s research centre in London.

Third Dimension was founded in 2024 by Tolga Kart and Bitor Sokolski, and previously raised seven million dollars in a seed round led by investors including Abstract Ventures, MVP, Soma Capital and Solar Capital. Although its technology was initially designed to accelerate video-game development by converting images and video into vivid 3-D worlds, the founders recognised early its potential for defense and autonomous-vehicle applications.The traditional simulation dilemma has always been the gap between game engines and the real world., where companies must inject massive amounts of synthetic data without bridging the physical gap, while the acquired technology uses artificial intelligence to match the physical world with extreme precision.

Revv bets on integrating these models with the flood of field data coming from its commercial and private fleets, which include trucks, multi-purpose sport utility vehicles and delivery buses equipped with activated external cameras. Philbin says the company receives more than 400,000 visual sequences each day from its customers’ fleets on the roads; the data are anonymised and aggregated and serve as the basis for training models. The essential difference lies in converting live real-world scenes into fully controllable, re-simulable environments; the system lets software engineers move camera angles, alter the behaviour of surrounding vehicles and pedestrians, and change lighting and weather conditions to test complex, rare edge cases without waiting for them to occur on actual pavement.

This direct investment in world models is not only tied to developing advanced driver-assist systems for consumer cars, but also fuels the core engine of autonomous-taxi vehicles that Revv plans to supply to Uber’s network. Competing in robot-axis fleet operations imposes safety requirements that exceed millions of kilometres of slow, costly physical testing.The ability to reconstruct a single real-world traffic scenario in hundreds of climatic and visual variationscuts years of software training and establishes a new benchmark for companies that relied exclusively on computer-graphics simulations separated from live street traffic.

What our coverage observes in this shift carries direct strategic implications for smart-transport and autonomous mobility plans in the Gulf and Egypt, especially for engineering teams, fleet operators and road authorities. In the region, where autonomous-taxi trials are accelerating and massive investments are poured into future-city infrastructure, operators often rely on standard digital-simulation systems or imported software that do not reflect field-environment specifics. Revv’s deal confirms that isolated driving algorithms lose value against world models fed with daily camera data; this means regional operators are now compelled to build live data-retrieval pipelines for their vehicles rather than rely on ready-made simulation solutions. If you are building or managing these systems, the required engineering skill is gradually moving away from merely tuning control software toward engineering visual world models capable of simulating sandstorms, harsh sunlight glare, and atypical traffic behaviours at the lowest possible computational cost.

The deal reveals that the race for autonomous vehicles has moved beyond a battle over on-board sensors to a battle over digitally representing reality. Accumulating millions of kilometres travelled is no longer sufficient if the system cannot decompose a scene and reconstruct it with maximum probability. When 400,000 daily frames are combined with an AI model that mirrors the physical world, the path becomes clear: safe autonomous driving will not be decided first on city streets, but in 3-D worlds generated from street details and tested in high-performance computing before a single wheel turns.