ANN ARBOR, MI
Dr. Ziqi Yu and his research team at the Toyota Research Institute of North America (TRINA) have made a breakthrough in broadband acoustic directional sensing of sirens using a resonator-based sensor, enhanced by deep learning. This innovation demonstrates the potential of artificial intelligence to extend the capabilities of physical sensing systems and promises significant improvements in autonomous vehicle safety.
Traditional resonator-based sensors face challenges such as limited bandwidth, performance degradation when compact, and susceptibility to environmental noise. This study shows that deep learning enhances directional sensing capabilities for both narrowband and broadband signals. The sensors, now more accurate by 20%-30%, utilize sound amplitude and phase better, extend operating frequencies beyond typical resonant ranges, and resist environmental noise; they have consistently outperformed their counterparts without resonators in a series of benchmark comparisons.
Deep learning can significantly extend the capability of compact resonator-based acoustic directional sensors by more effectively employing the acoustic features processed by the resonant effect. This mechanism may be extended to miniaturized sensing systems based on different wave physics that exhibit resonant phenomena.
These advancements offer cost-effective alternatives to optical-based directional sensing systems in vehicles, .e.g., for the application of emergency vehicle detection, showcasing superior sensing capabilities due to the synergy between deep learning and resonant effects. This paves the way for intelligent and safe future mobility solutions.
Despite the promising results, the study was conducted with neural networks trained on local computers and did not consider moving sound sources or multiple simultaneous noises. Future work will focus on developing smaller, on-chip sensors, methods for on-chip neural network training.
For a detailed information, please refer to the published paper in Scientific Reports.