Collaborative Approach to Sensor Perception Modelling (SENSI)

Preprint, 2025

Automated driving systems (ADS) rely on perception pipelines that are exposed to multiple sources of uncertainty: sensor limitations, environmental degradation, noise and error in human-generated labels, and perception model errors. Most simulation and evaluation frameworks still assume ideal perception, which creates a mismatch between virtual assessments and real-world performance. The SENSI project develops an open, data-driven stochastic perception model that captures these uncertainties.