![nueral network radar nueral network radar](https://www.skyradar.com/hubfs/Images/Blogs/Radar-Ai-Input-Hidden-Layer-output.png)
The neuromorphic computing field (Roy et al., 2019) presents an attractive alternative to overcome the previously described challenges. ( 2018) Copyright 2018 American Chemical Society. Reprinted with permission from Gawron et al. (B) Sources of added energy consumption on a medium automated vehicle system on an electric vehicle prototype. (A) Exemplary sensor setup of an automated vehicle prototype. While the energy per operation in CPUs and GPUs decreases for smaller semiconductor manufacturing processes, researchers see an asymptotic efficiency wall that is slowly approached in the next years (Marr et al., 2013): Therefore, alternative approaches regarding hardware and algorithms are demanded that fulfill both the efficiency and safety requirements for autonomous vehicles. Furthermore, in electric vehicles high processing demands can significantly reduce the travel range.
![nueral network radar nueral network radar](https://static.seekingalpha.com/uploads/2013/1/5731571_13593188717732_1.png)
This combination of increasing in-vehicle deployment of modern and power-hungry machine learning approaches rich and redundant sensor setups and limited on-board energy resources poses significant challenges on the realization of automated vehicles: Already today, a significant amount of energy in automated vehicle prototypes is dedicated to computing (Gawron et al., 2018, see also Figure 1B). On the other hand, automated vehicle prototypes are typically equipped with a rich setup of various sensor units (Aeberhard et al., 2015, see also Figure 1A) to ensure a sufficient coverage of the vehicle's surroundings as well as safety through sensor redundancy. Therefore, the use of such powerful learning approaches in automated vehicle functions and components is likely to increase in the near future. One key aspect of this development is the success of modern machine learning approaches over the past decade, particularly deep learning by achieving remarkable results on several tasks necessary for fully automated driving, such as traffic sign recognition (Ciresan et al., 2012), semantic segmentation (Badrinarayanan et al., 2015), 2D and 3D object detection (Zhou et al., 2019 Yin et al., 2020), and behavior prediction of other traffic participants (Deo and Trivedi, 2018). This study proves the general applicability of SNNs for automotive radar processing and sustains the prospect of energy-efficient realizations in automated vehicles.Īutomated driving is currently a very appealing area of research continuously drawing attention from academic and industrial research groups alike. Finally, we discuss the specific requirements and challenges for SNN-based radar processing on neuromorphic hardware. Instead, our SNN for target classification achieves an accuracy close to a reference artificial neural network while requiring 200 times less operations. For radar target detection, an SNN with temporal coding is competitive to the conventional approach at a low compute overhead. We provide SNN examples for two processing steps and evaluate their accuracy and computational efficiency. In this article we perform a step-by-step analysis of automotive radar processing and argue how spiking neural networks could replace or complement the conventional processing. One promising approach for energy-efficient signal processing is the usage of brain-inspired spiking neural networks (SNNs) implemented on neuromorphic hardware. However, the rising number of transmitters and receivers for obtaining a higher angular resolution increases the cost for digital signal processing. Frequency-modulated continuous wave radar sensors play an essential role for assisted and autonomous driving as they are robust under all weather and light conditions.