Rivian buys world model startup to accelerate self-driving ambitions

Rivian acquired Third Dimension, a U.K. startup founded in 2024 by gaming and autonomous-vehicle veterans. Third Dimension uses AI to turn images and videos into lifelike 3D worlds. Rivian wants this technology for driver-assistance development and future robotaxi systems supplied to Uber.

The startup’s 15-person team will be absorbed into Rivian’s autonomy and AI team, led by James Philbin. Most team members are based in the U.K. Their technology was originally built to speed video-game development, but the founders also identified uses in defense and autonomous vehicles.

The deal’s financial terms were not disclosed. Rivian also sees the acquisition as a chance to expand its AI hub in London. Under Rivian, Third Dimension’s team will focus on automated driving rather than serving several industries at once.

An AI-generated 3D world model is a computer-built representation of places, objects, roads, and surrounding conditions. It aims to behave and look more like the physical world than a purely synthetic game environment. Rivian wants such models to improve automated-driving development.

Third Dimension starts with images and videos. Its AI learns visual patterns in that material and creates a lifelike 3D world that represents what the cameras captured. The article does not explain the exact reconstruction pipeline, such as how depth or object movement is calculated. It does establish that the system matches the physical world.

This matters because driving systems need realistic places to train and test. Rivian can combine the startup’s technology with its own large collection of road data. That could produce simulations of difficult conditions without waiting for those conditions to appear during ordinary driving.

Rivian collects more than 400,000 driving sequences uploaded from customer fleets each day. The data comes from tens of thousands of vehicles on the road. Rivian uses it to build training models for driving, making the fleet a major source of real-world examples.

Contributing vehicles include R1S pickup trucks, R1S SUVs, delivery vans, and R2 SUVs. The article says the data comes from vehicles whose owners have enabled external camera sharing. This means the collection depends on customer participation rather than every vehicle automatically contributing footage.

Rivian says the data is aggregated and anonymized. That data can be paired with Third Dimension’s AI technology to create more realistic virtual driving environments. The article does not state how much data comes from each vehicle type, but it highlights the scale: hundreds of thousands of sequences arrive every day.

Autonomous-vehicle developers use simulation to train and validate self-driving systems. A simulator can expose a system to many situations in a controlled virtual environment. This is important because real roads may not provide enough examples of every difficult condition during normal testing.

Historically, simulators were built with rendering or video-game engines. They often relied on synthetic data, so their virtual environments could look realistic without fully matching real-world behavior and appearance. According to Third Dimension’s founder Tolga Kart, that mismatch creates a gap between simulation and reality.

The usual workaround is to feed simulators increasingly large amounts of data. More data may help narrow the quality gap, but it does not automatically make a synthetic environment identical to the physical world. Rivian hopes Third Dimension’s AI can create worlds that match real conditions more closely.

Combining Third Dimension’s technology with Rivian’s customer-vehicle data could create a simulator grounded in real driving. The technology is designed to match the physical world, while Rivian contributes footage and sequences gathered from vehicles on the road. Together, they could make virtual training and testing more realistic.

For example, Rivian could use its fleet data to represent roads, vehicles, and surrounding conditions inside a 3D environment. The simulator could then reproduce complex situations for automated-driving systems. The article does not identify specific recreated scenarios, but it describes the goal as simulating difficult conditions.

James Philbin said this approach could avoid waiting for rare real-world edge cases to occur. It could accelerate development of Rivian’s advanced driver-assistance features and the self-driving system for robotaxis it plans to supply to Uber. The acquisition therefore links data collection, simulation, and future deployment plans.

Rare driving situations are difficult to study through ordinary road use. A dangerous hazard or near-crash may happen infrequently, and developers cannot schedule it safely for testing. Simulation offers a controlled alternative by placing the driving system in a virtual version of a challenging situation.

With Third Dimension’s technology, Rivian could use real-world images, videos, and fleet data to build lifelike environments. Developers could then test how a system responds to complex conditions and repeat those tests. The article does not specify the exact hazards the simulator will contain, but it says the technology can simulate complex driving conditions.

This could shorten development time and improve validation. Instead of waiting for rare cases to appear naturally, Rivian could study them virtually as soon as suitable data and models are available. The approach is intended for advanced driver-assistance features and the self-driving systems planned for Uber robotaxis.

Advanced driver-assistance systems, or ADAS, support a human driver with selected driving functions. Fully autonomous driving aims to control the driving task without that same level of human involvement. This distinction is general automotive knowledge; the article mentions both Rivian’s driver-assistance features and planned self-driving robotaxis but does not define them.

Both kinds of systems must learn from driving data and be checked in many conditions. Training helps a system recognize roads, vehicles, and hazards. Validation tests whether it responds as intended, including in situations that may be difficult or uncommon. Simulation can provide those situations in a controlled setting.

For Rivian, Third Dimension’s technology could support both goals. Its AI-generated worlds may make training and testing more realistic, while Rivian’s fleet supplies real-world material. The article connects this work to Rivian’s current driver-assistance development and its planned self-driving systems for Uber robotaxis.