Sensor fusion algorithm

 Kalman filter

Kalman filter is typical.

The core of the algorithm is to set a set of "belief" factors for each sensor. At each moment, the sensor data from the last moment will be used for statistics to improve the guess (self-adding), and the quality of the sensor will also be judged. In the comparison between the predicted value and the measured value of the sensor, an excellent value will be estimated and output.

This means that if a sensor always gives a good and consistent value and starts telling you something unlikely, the sensor's credibility level will decrease in a few milliseconds until it starts to make sense again.

This is better than simple averaging or voting, because Kalman filter can deal with the situation that most sensors are temporarily out of order. As long as one can keep good reason, it can make the robot through the dark moment.

Kalman filter is an application of more general concepts of Markov chain and Bayesian reasoning, which is a mathematical system that iteratively improves their guesses by using evidence. These tools are tools used to help science itself test ideas (which are also the basis of what we call "statistical significance").

Therefore, it can be said poetically that some sensor fusion systems are expressing the essence of science at a speed of 1000 times per second.

Kalman filters have been used in orbital stations of space satellites for decades. Because modern microcontrollers can run the algorithm in real time, they are becoming more and more popular in robotics.