Possibilistic filtering for reliable robot localization

Uncertainty in engineering applications is commonly addressed within a probabilistic framework that assumes an underlying aleatory nature. However, in many practical scenarios, uncertainty can arise from incomplete knowledge rather than inherent randomness, reflecting epistemic rather than purely stochastic effects. Classical probabilistic approaches do not explicitly distinguish between these types of uncertainty, which can lead to overly confident or diluted probability assessments. This motivates the use of more expressive frameworks for uncertainty quantification. Possibility theory, as a theory of imprecise probabilities, provides a unified and computationally efficient framework for representing both epistemic and aleatory uncertainty. In particular, possibilistic filtering techniques have recently been developed for dynamic state estimation problems. In this contribution, we apply a particle-based possibilistic filtering approach to a real-world robot localization problem. The considered setup involves significant epistemic uncertainty arising from ambiguous landmark configurations and measurement limitations. The results demonstrate that the proposed method yields robust state estimates that reflect the underlying uncertainty. Furthermore, we introduce modifications to the original filter implementation that substantially reduce the computational cost.