Occupancy grid mapping (OGM) is a fundamental building block in probabilistic, spatial world models that support safe navigation and decision-making in both autonomous vehicles and mobile robots. This tutorial provides a blend of OGM theory and practice, connecting Bayesian estimation principles with modern autonomy stacks.
We begin with classical formulations: inverse sensor and Bayesian models, and then progress to learning based OGM. We then discuss extensions to dynamic environments (moving objects, temporal filtering, and tracking in grid space) and 3D/semantic grids for richer scene understanding. The session emphasizes impact of anomalies in real-world sensor data from modalities like LiDAR and radar used in OGM, and robustness in adverse conditions.
Participants will learn how occupancy grids interface with downstream modules and how to evaluate mapping quality using metrics aligned with real-time constraints and safety requirements. Overall, the tutorial offers both conceptual insights and practical guidance for building robust occupancy-based perception systems in real-world settings.