3 ms·
Systems designed to prevent cars hitting pedestrians are not designed for extremely low speed environments where these noise makers are most effective. Noise ma
by tialaramex 7y ago
Systems designed to prevent cars hitting pedestrians are not designed for extremely low speed environments where these noise makers are most effective. Noise makers make no difference at highway speed, and in examples I've seen they aren't required at those speeds anyway. But they fix electric cars to not "sneak up" on pedestrians in these low speed environments.
At 3mph your car can very easily knock me down and crush me if I don't hear it coming, but designing software to recognise that it's going to collide with me, and not pass harmlessly a few metres away (a routine event you don't want the AI to freak out over) is really hard.
These low speeds are correlated with environments where it's much harder for a machine to intuit what's going on. Consider a typical open air car park next to a Walmart. A pedestrian is in view, then vanishes, the driver selects reverse - are we going to hit the pedestrian? Where did they go? The driver knows, because they have relevant experience, that the pedestrian is between two other parked cars, and won't be struck by reversing, but how can our AI figure that's where they went?
- zaroth 7y agoDefinitely this has nothing at all to do with highway speeds. We’re talking parking lots mainly. I don’t see the need for intuition. What’s the stopping distance at 3mph? Less than 1 foot with full pre-charged braking I’m sure. An effective system should be able to see a pedestrian 6-12” from the car in the direction of travel, and stop on a dime. At these speeds a false positive isn’t itself a safety risk, although of course you would seek to minimize them. It goes without saying the system will not be perfect. Neither are noise makers. I see no reason why such a system couldn’t be made that is more effective than fake engine noise.