# Bayesian Learning of Occupancy Grids

**Authors:** Christopher Robbiano, Edwin K. P. Chong, Louis L. Scharf, Mahmood R. Azimi-Sadjadi, Ali Pezeshki, Saeid Ahmadinia  
**arXiv:** [1911.07915](https://arxiv.org/abs/1911.07915)  
**Date:** 2019-11-18  
**Summary page:** https://www.summarizepaper.com/en/paper/1911.07915v1/

## Key points

- The paper introduces a novel framework for updating occupancy probabilities in two-dimensional grids based on sensor measurements.
- Traditional occupancy grids estimate the likelihood of each cell being occupied over time using disjoint cells, but the proposed framework utilizes binary asymmetric channel models to account for observation errors.
- The authors analyze binary-valued measurement vectors generated by physical layer detectors in various sensory systems and demonstrate that their Bayesian approach outperforms classical formulations for occupancy grids.
- The new framework achieves lower false alarm and miss rates while requiring fewer observations to accurately estimate occupancy probabilities in the surrounding area.
- This research contributes to advancing techniques for updating occupancy grids with improved accuracy and efficiency by leveraging Bayesian principles and accounting for error-prone observations in sensor data analysis.

## Summary

The paper "Bayesian Learning of Occupancy Grids" introduces a novel framework for updating occupancy probabilities in two-dimensional grids based on sensor measurements. Traditional occupancy grids aim to estimate the likelihood of each cell being occupied over time using disjoint cells. However, the proposed framework departs from this assumption and utilizes binary asymmetric channel models to account for observation errors. By analyzing binary-valued measurement vectors generated by physical layer detectors in imaging, radar, sonar, or other sensory systems, the authors demonstrate that their Bayesian approach outperforms classical formulations for occupancy grids. The new framework achieves lower false alarm and miss rates while requiring fewer observations to accurately estimate occupancy probabilities in the surrounding area. This research contributes to advancing techniques for updating occupancy grids with improved accuracy and efficiency by leveraging Bayesian principles and accounting for error-prone observations in sensor data analysis.