There and Back Again: Learning to Simulate Radar Data for Real-World Applications
This work is significant for researchers and developers working on data-driven radar processing, as it offers a way to accelerate development by generating realistic synthetic data, addressing the challenge of complex radar image formation.
This paper tackles the problem of simulating realistic radar data by learning a radar sensor model that synthesizes radar observations from simulated elevation maps. The proposed method enables a downstream segmentation model trained purely on simulated data to achieve performance within four percentage points of a model trained entirely on real data.
Simulating realistic radar data has the potential to significantly accelerate the development of data-driven approaches to radar processing. However, it is fraught with difficulty due to the notoriously complex image formation process. Here we propose to learn a radar sensor model capable of synthesising faithful radar observations based on simulated elevation maps. In particular, we adopt an adversarial approach to learning a forward sensor model from unaligned radar examples. In addition, modelling the backward model encourages the output to remain aligned to the world state through a cyclical consistency criterion. The backward model is further constrained to predict elevation maps from real radar data that are grounded by partial measurements obtained from corresponding lidar scans. Both models are trained in a joint optimisation. We demonstrate the efficacy of our approach by evaluating a down-stream segmentation model trained purely on simulated data in a real-world deployment. This achieves performance within four percentage points of the same model trained entirely on real data.