This is a pretty interesting paradigm shift for AutoPilot. Currently the system is comprised of three main sensors:
The primary sensor for AutoPilot is currently the camera. It picks up the lanes, identifies cars, objects, etc. Radar spaces the car out between itself and the vehicle in front of it and the sonar sensors detect objects in a 16 foot bubble surrounding the car. So if someone is merging into they see that. Otherwise they are a distant third as far as information goes and you can actually engage Autopilot without them.
The problem with the camera is at sunset when theres light being shined directly into it, or on rainy and foggy days. The bigger problem is the camera just isn't as good at determining what things are. On the flip side, radar can see through fog, rain, sun, etc but throws up a lot of false negatives. Elon said in his notes that radar might perceive the curved bottom of a coke can as being a truck which is why it can't be relied upon for heavy lifting... Until now.
The Change
As of the next firmware version Tesla is now giving the radar sensor equal decision making power in auto-pilot mode. There is no need for new hardware. It is purely a software change. Tesla says they've fixed the false negative problem two ways:
1. They have a new driver that lets them get a higher resolution "picture" from the existing sensor.
2. This is the cool one: They will use the existing fleet of Tesla cars to determine whether or not an object detected by radar is indeed an object or just something else like a street sign.
The way it works is that as existing cars pass a point on the road where the radar determines there is a threat, the car will record how the person driving the vehicle responds to the threat versus how the car would have responded were it relying on radar alone. So if I pass a street sign and the radar sees that as an object in the road, as I pass by the sign without braking that specific area will be whitelisted so that other Teslas know when they pass that specific in that specific area with that specific radar signature that it's not a threat and the car can ignore the false positive. Presumably until there is enough data the cars will work the way they work now by acknowledging the threat and then referencing the camera to determine whether or not it's actionable.
The advantage here is that Elon says the new system will bounce under or through other cars so that AutoPilot now not only sees the car in front of you but also the car in front of that one. So if the person two cars in front of you slams on their brakes and the guy in front of you doesn't see it, your car still will see it.
The other big issue is that it addresses situations that the current system is apparently designed to ignore. For instance, the situation where the guy watching Harry Potter was cut off by a big rig and he crashed into it's trailer, the existing system sees that with the camera and assumes it's looking at a highway overpass sign. The updated system would actually see that as a threat and stop the car.
It's a pretty intuitive way of making the best of the sensors you already have as well improving the system.
Also, since this forum isn't about cars any more, Hillary sucks and I think every American should be disarmed. Also I like boats and 15 year old GM trucks. That should get things started.
- Sonar
- Radar
- Video Camera
The primary sensor for AutoPilot is currently the camera. It picks up the lanes, identifies cars, objects, etc. Radar spaces the car out between itself and the vehicle in front of it and the sonar sensors detect objects in a 16 foot bubble surrounding the car. So if someone is merging into they see that. Otherwise they are a distant third as far as information goes and you can actually engage Autopilot without them.
The problem with the camera is at sunset when theres light being shined directly into it, or on rainy and foggy days. The bigger problem is the camera just isn't as good at determining what things are. On the flip side, radar can see through fog, rain, sun, etc but throws up a lot of false negatives. Elon said in his notes that radar might perceive the curved bottom of a coke can as being a truck which is why it can't be relied upon for heavy lifting... Until now.
The Change
As of the next firmware version Tesla is now giving the radar sensor equal decision making power in auto-pilot mode. There is no need for new hardware. It is purely a software change. Tesla says they've fixed the false negative problem two ways:
1. They have a new driver that lets them get a higher resolution "picture" from the existing sensor.
2. This is the cool one: They will use the existing fleet of Tesla cars to determine whether or not an object detected by radar is indeed an object or just something else like a street sign.
The way it works is that as existing cars pass a point on the road where the radar determines there is a threat, the car will record how the person driving the vehicle responds to the threat versus how the car would have responded were it relying on radar alone. So if I pass a street sign and the radar sees that as an object in the road, as I pass by the sign without braking that specific area will be whitelisted so that other Teslas know when they pass that specific in that specific area with that specific radar signature that it's not a threat and the car can ignore the false positive. Presumably until there is enough data the cars will work the way they work now by acknowledging the threat and then referencing the camera to determine whether or not it's actionable.
The advantage here is that Elon says the new system will bounce under or through other cars so that AutoPilot now not only sees the car in front of you but also the car in front of that one. So if the person two cars in front of you slams on their brakes and the guy in front of you doesn't see it, your car still will see it.
The other big issue is that it addresses situations that the current system is apparently designed to ignore. For instance, the situation where the guy watching Harry Potter was cut off by a big rig and he crashed into it's trailer, the existing system sees that with the camera and assumes it's looking at a highway overpass sign. The updated system would actually see that as a threat and stop the car.
It's a pretty intuitive way of making the best of the sensors you already have as well improving the system.
Also, since this forum isn't about cars any more, Hillary sucks and I think every American should be disarmed. Also I like boats and 15 year old GM trucks. That should get things started.