Is there any official information on why nearby cars jump around on the Tesla display?
As an engineer, I speculate that it's due to noisy sensors, primarily the noisy distance estimates of camera and ultrasonic sensors. I notice that the cars in front of my car are rendered much more stably, likely due to the accuracy of the forward-looking radar. But cameras have a much harder time estimating sizes and distances. And ultrasonic sensors are just noisy because of all of the signal reflections on multiple paths.
But that's not the complete story because is doesn't explain why the nearby cars are rendered in a way that includes the sensor noise. They could use filtering techniques to essentially make the best guess of each car's location despite the noise. Or they could add in physics constraints that say, for example, that cars don't jump 3 feet in a fraction of a second, don't bounce out of their lanes, and don't overlap with other cars.
The problem with these techniques is that they'll introduce lag or other errors.
Lag: The noisy sensor data is still being input to the physics model. It's just that the model updates more sensibly as new data arrives. But that means that it will take more data around a given spot for the model to update to that location: That's a lag. So that would make the nearby cars to appear sluggish. When the cars start accelerating from a stoplight, for example, the display would take some time to show them moving. That could be dangerous if a driver used the display to decide to change lanes.
Other errors: Because of the noisy data, you can't be sure where to apply the constraints. If you say, for example, that two cars shouldn't overlap, which car should you move? One, or the other, or both? Each adds information not present in the sensor data. That potentially introduces additional errors, the opposite of the original intention.
Tough problem. The above may provide some reasons why the cars dance around the Tesla screen. Other ideas?
As an engineer, I speculate that it's due to noisy sensors, primarily the noisy distance estimates of camera and ultrasonic sensors. I notice that the cars in front of my car are rendered much more stably, likely due to the accuracy of the forward-looking radar. But cameras have a much harder time estimating sizes and distances. And ultrasonic sensors are just noisy because of all of the signal reflections on multiple paths.
But that's not the complete story because is doesn't explain why the nearby cars are rendered in a way that includes the sensor noise. They could use filtering techniques to essentially make the best guess of each car's location despite the noise. Or they could add in physics constraints that say, for example, that cars don't jump 3 feet in a fraction of a second, don't bounce out of their lanes, and don't overlap with other cars.
The problem with these techniques is that they'll introduce lag or other errors.
Lag: The noisy sensor data is still being input to the physics model. It's just that the model updates more sensibly as new data arrives. But that means that it will take more data around a given spot for the model to update to that location: That's a lag. So that would make the nearby cars to appear sluggish. When the cars start accelerating from a stoplight, for example, the display would take some time to show them moving. That could be dangerous if a driver used the display to decide to change lanes.
Other errors: Because of the noisy data, you can't be sure where to apply the constraints. If you say, for example, that two cars shouldn't overlap, which car should you move? One, or the other, or both? Each adds information not present in the sensor data. That potentially introduces additional errors, the opposite of the original intention.
Tough problem. The above may provide some reasons why the cars dance around the Tesla screen. Other ideas?
