There and Back Again: Learning to Simulate Radar Data for Real-World Applications

Abstract

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.

Radar Occupancy Prediction with Lidar Supervision while Preserving Long-Range Sensing and Penetrating Capabilities

“…The top row shows the radar occupancy prediction training process. The middle and bottom rows illustrate the main issues of radar-tolidar training and show the comparison between our and previous works's results [5], [6]. The middle row also shows the proposed method can successfully translate radar to lidar while preserving long-range sensing.…”

“…But these methods are troublesome to remove multipath reflection with high power returns. Recently, data-driven methods have been proven to be effective in applications for radar noise filtering [4], [5], [7], [14], [15]. In [14], the authors use radar submap as training ground truth to obtain noiseless radar.…”

“…The radar-to-lidar training achieved by generative adversarial network (GAN) [16] is proposed in [6]. The idea has been explored to height estimation in [5]. However, by doing this, the sensing range of the result was restricted by the lidar.…”