Self-localisation through System Identification

We analyse the relationship between a mobile robot’s perception, motion and position (i.e. we address the problem of sensor-based self-localisation), by constructing transparent, linear polynomial models, using an ARMAX process which expresses the interdependency between motion, perception and location. Our models show that in sensor-based localisation there is a clear and measurable trade-off between the amount of sensor information used and the length of time over which the information is observed. We also demonstrate that in cases of repetitive, predictable behaviours the best method of self-localisation is to take previous position estimates into account.