Radar Sensor Model for Three-Dimensional Map Building

Abstract

Radar offers advantages as a robotic perception modality because it is not as vulnerable to the vacuum, dust, fog, rain, snow and light conditions found in construction, mining, agricultural and planetary-exploration environments. However radar has shortcomings such as a large footprint, sidelobes, specularity effects and limited range resolution-all of which result in poor environment maps. Evidence grids are a flexible and powerful probabilistic method for fusing multiple sensor observations. Sensor models exist for interpreting the range readings of sonar, laser and stereo. However, these existing sensor models do not work with radar because it provides amplitude values for many points downrange. In addition, radar has significant echo signal-to-noise variations between observations as well as limited downrange resolution. This paper presents the development of a radar sensor model, which can fuse amplitude-vector sensor data into an evidence grid. A study of radar phenomena and of frequency-modulated continuous-wave signal processing suggests rules for signal interpretation. The sensor model uses these interpretation rules and captures the volumetric beam geometry. The results include a three-dimensional map of an outdoor scene. This work is a step towards building high fidelity maps to be used in mobile robot navigation, obstacle avoidance and tool deployment under all visibility conditions.