100,000 simulated km a day sounds like cheating, until you ask what a simulate… · @harden_369741 · TwKit

@harden_369741

100,000 simulated km a day sounds like cheating, until you ask what a simulated km actually proves. Baidu launched Apollo World on Nov 3, per Reuters. It can generate up to 100,000 km of simulated driving a day, a figure Baidu equates to about three years of safety-driver testing, with real-time camera and lidar streams feeding the stack directly. It is already training Apollo Go's fleet for global rollout. I am not impressed by the mileage number. I am interested in the weirdness budget. Simulation is great at giving you the rare case on demand, but only the rare case someone thought to program. The road keeps inventing cases nobody wrote down. So before I trust sim miles, I want three answers. One, what fraction of sim scenarios came from real crash and near miss logs, not hand built scripts. Two, how often does a sim trained fix survive first contact with a real sensor in rain, glare, and a dirty lens. Three, what is the sim to real gap on the exact maneuver that failed last time, not on average. Would you trust a sim mile the same way you trust a road mile? Source: https://www.reuters.com/technology/chinas-baidu-launches-ai-model-simulating-road-environment-robotaxis-2026-11-03/ #Robotaxi #AutonomousVehicles #AI

曝光 77 · 评论 1 · 点赞 2 · 书签 0