Abstract:Automotive Multiple-Input Multiple-Output (MIMO) radars have gained significant traction in Advanced Driver Assistance Systems (ADAS) and Autonomous Vehicles (AV) due to their cost-effectiveness, resilience to challenging operating conditions, and extended detection range. To fully leverage the advantages of MIMO radars, it is crucial to develop an Angle of Arrival (AOA) algorithm that delivers high performance with reasonable computational workload. This work introduces AAETR (Angle of Arrival Estimation with TRansformer) for high performance gridless AOA estimation. Comprehensive evaluations across various signal-to-noise ratios (SNRs) and multi-target scenarios demonstrate AAETR's superior performance compared to super resolution AOA algorithms such as Iterative Adaptive Approach (IAA). The proposed architecture features efficient, scalable, sparse and gridless angle-finding capability, overcoming the issues of high computational cost and straddling loss in SNR associated with grid-based IAA. AAETR requires fewer tunable hyper-parameters and is end-to-end trainable in a deep learning radar perception pipeline. When trained on large-scale simulated datasets then evaluated on real dataset, AAETR exhibits remarkable zero-shot sim-to-real transferability and emergent sidelobe suppression capability. This highlights the effectiveness of the proposed approach and its potential as a drop-in module in practical systems.
Abstract:There is a renewed interest in radar sensors in the autonomous driving industry. As a relatively mature technology, radars have seen steady improvement over the last few years, making them an appealing alternative or complement to the commonly used LiDARs. An emerging trend is to leverage rich, low-level radar data for perception. In this work we push this trend to the extreme -- we propose a method to perform end-to-end learning on the raw radar analog-to-digital (ADC) data. Specifically, we design a learnable signal processing module inside the neural network, and a pre-training method guided by traditional signal processing algorithms. Experiment results corroborate the overall efficacy of the end-to-end learning method, while an ablation study validates the effectiveness of our individual innovations.