Cumulative error estimation from noisy relative measurements

Odometry is important for autonomous vehicle in scenarios where GPS is either unavailable or only intermittently available. However, in a large scale environment, it often generalizes unbounded cumulative error when the vehicle unconsciously moves. This paper analyzes how the cumulative error grows according to the noisy relative measurements. An unbounded drift model is proposed to represent the cumulative error, where its probability distribution is described by the corresponding expectation and variance. Compared to other approaches, it presents a recursive cumulative error expression in absence of the true positions, which has great potentials in various domains, e. g. path planning, odmetry based localization. Both experiments and cases are conducted to not only verify the accuracy of the proposed model, but also illustrate the potentials in related domains.