Self-Driving Wheelchairs and Scooters Just Got Much Safer

Small self-driving vehicles are having a moment. Motorized wheelchairs, scooters and e-bikes could give people who can't drive a cheap, independent way to get around. But the software running most self-driving cars today was built for cars and trucks. It assumes these little vehicles move like simple toy cars on rails — ignoring tire slip, uneven friction and the fact that a bicycle and a wheelchair lean and turn completely differently. That makes the ride jerky, unpredictable and sometimes unsafe around pedestrians.

A research team from the University of Maryland took a different approach. They built two new mathematical models of how these vehicles really move, then made them "differentiable" — a fancy word meaning the AI can trace its mistakes backward and smoothly adjust, the way a cook tweaks a recipe after tasting it. Before, the software could only guess and check. Now it can learn from every trip. One model handles bicycles and scooters; the other covers a whole family of small vehicles, from wheelchairs to motorcycles.

They then tested both in a computer simulation of crowded sidewalks. The vehicles had to plan their own routes around moving pedestrians. The tire-based model beat the simpler one clearly: 55% fewer collisions and 75% fewer moments that would have felt jarring or unsafe for a rider, based on the bicycle test case. It also reproduced real recorded rides more accurately.

The catch: all of this happened in simulation, not on actual streets. No real wheelchair or scooter was driven, and "discomfort" is a number the researchers calculated, not a report from an actual rider. The work is a promising building block, not a product you can buy. Still, it points toward a future where a wheelchair could navigate a busy sidewalk as confidently as a careful human.