Waymo develops new benchmark model to compare robotaxis with human drivers
Original: Waymo says it built a better benchmark for comparing robotaxis to humans
Why This Matters
Critical for autonomous vehicle safety validation as companies scale operations
Waymo and TU Delft published research in Nature Communications on a new computer model using active inference theory to better simulate human driving behavior. The Reference Driver model can reproduce pre-crash behavior, not just last-second reactions like previous models.
Waymo created a new computer model called the Reference Driver in partnership with TU Delft to more accurately compare autonomous driving performance with human drivers. Published in Nature Communications, the model uses active inference theory - the concept that drivers constantly imagine possible futures and take actions toward the safest outcome. Unlike previous models that focused on last-second reactive maneuvers, the Reference Driver can simulate the entire lead-up to crashes and the internal 'surprise' drivers feel during conflicts. The development comes as Waymo faces increased scrutiny following a January incident where its robotaxi struck a child in Santa Monica at 6 mph after decelerating from 17 mph. The company previously used older models to claim a human would have hit at 14 mph. The new model can be applied to thousands of scenarios and adapted for various road user behaviors beyond collision avoidance.