3 ms·
The parent said weighted average. The weight can be picked cleverly: it does not have to be the 0.5 of your example, and it doesn’t have to be constant. For ex
by robotresearcher 5y ago
The parent said weighted average. The weight can be picked cleverly: it does not have to be the 0.5 of your example, and it doesn’t have to be constant.
For example, the Kalman Filter tells you the weight to apply to each new sample and the last estimate to make a new estimate. These weights are optimal if the measurement errors have a Gaussian distribution with zero mean. In many practical systems they are usably good even when that assumption does not perfectly hold.
In your example the new measurement that was less consistent with the last fused estimate would be downweighted by a KF.
- sandworm101 5y ago>> they are usably good even when that assumption does not perfectly hold. And when deployed against a jamming target that it well aware of gaussian distributions and how to game them? When deployed against a jammer that is very good at pretending to not be a jammer? An aware jammer will feed you good results, just enough to convince your gaussian distribution (capture) and them move it off of onto a false path (seduction) resulting in you following a false signal (miss) or cause you to fall back into search mode (delay=miss). The game is different when the thing you are trying to measure is actively trying to deceive you.
- robotresearcher 5y agoYou are right. Countermeasures would deliberately violate the assumptions a KF or other estimator makes, to degrade its ability to track. Now you have an arms race between countermeasures and the ability to model and defeat them.