Imagine you are in a self-driving taxi, but the ride is a little too aggressive. Maybe the car is taking turns too sharply, or you are running late and wish it would pick up the pace. Right now, most autonomous vehicles are locked into a single driving style set by engineers before they ever leave the parking lot. You have no way to give feedback.
Researchers at Delft University of Technology have developed a way to let passengers influence these driving habits using everyday language. Instead of a driver manual or a complicated dashboard, you could simply tell the car, I am running late or I am feeling dizzy. The car uses a large language model — a system designed to understand and generate human-like text — to interpret your request and translate it into technical settings the car understands. Before it changes anything, the system explains its plan in plain English and asks for your confirmation.
Giving the car a personality
To understand how this works, think of the car as having two different systems. One is the navigator, which handles safety. It is a strictly controlled, logical piece of software that calculates paths, watches for other cars, and obeys traffic laws. This part of the brain is never allowed to be creative or experimental because it must be predictable and safe.
The large language model acts as a separate, secondary layer sitting on top. It does not control the car directly. Instead, when you make a request, the model acts like a translator. It takes your subjective request—like wanting a smoother ride—and converts it into numerical scores for the navigator. If you ask for comfort, the model tells the navigator to prioritize gentle acceleration and slow turns. If you ask to go faster, it nudges those specific safety-bound settings closer to the limit. The navigator then carries out your request, but only within the rigid, pre-programmed safety boundaries that prevent it from doing anything dangerous. By keeping the language model separate from the actual steering and braking, the car can be conversational without risking safety.
This approach addresses one of the biggest hurdles for self-driving cars: trust. If a car feels like a reckless driver, humans feel anxious. By allowing us to interact with the vehicle in a way that feels like giving directions to a human taxi driver, companies might make autonomous travel feel more natural and responsive to our needs. It shifts the role of the AI from being a rigid machine that only follows strict rules, to a service that respects the person sitting in the backseat.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy