Proving the Safety of Motion Estimation Algorithms: a Confidence Interval Approach

This paper presents innovative algorithms designed for proving the integrity of parameter estimation. These algorithms rely on an unusual sensor modeling, delivering measurements as confidence intervals (with an associated confidence level) rather than as point estimates, as usually considered. Suitable combinations of measurements provided by several sensors may reach extremely high confidence levels for the estimate, and may thus serve estimation problems where safety is of uttermost importance. One typical application is motion estimation for automated guided transport (railways for instance), but our confidence-interval based algorithms are general enough to apply for other safety-demanding applications