Even noisy data from AprilTag localization can be used to heavily improve robot odometry over just tracking the velocity of the robot's wheels.
In this case the previous position estimate is first improved by moving it 70% of the way to the latest velocity-based prediction of the robot's location, and then the estimate is further improved by moving it either 10% or 50% of the way towards each AprilTag-reported location. We update the estimate with only 10% if the robot is ~stopped, in order to smooth out the noise of the AprilTag, but move 50% of the way there when moving in order to reduce signal lag based on movement.
15 сен 2024