Work

Heterogeneous multimodal driver assistance using a Jetson Nano–ESP32 architecture

Year: 2026
Type: article
Abstract: Low-cost embedded advanced driver-assistance systems must combine heterogeneous perception and control functions within limited compute and memory budgets. We present a five-function robotic ADAS prototype that partitions vision and RGB-D association to an NVIDIA Jetson Nano while assigning ultrasonic blind-spot sensing, warnings and actuation to an ESP32. A single Astra Pro Plus RGB-D camera supplies forward color and depth for vehicle, traffic-sign, speed-bump and lane processing; two lateral HC-SR04 sensors provide the blind-spot path. The final 44-epoch ACC validation on 1,126 images (3,517 vehicle instances) produced precision 0.869, recall 0.823, [email protected] = 0.889 and [email protected]:0.95 = 0.684. The selected traffic-sign YOLOv5s configuration recorded 95.9% precision and 92.5% recall. For speed-bump detection, YOLOv5s recorded 94.2% precision, 91% recall and [email protected] = 0.843 while using 91.7% fewer parameters and requiring 91.6% less storage than YOLOv5x. Lane and blind-spot tests provide functional demonstrations of sensing-to-warning and sensing-to-actuation paths. The study establishes architectural feasibility and subsystem-level experimental validation on a low-cost robotic prototype, not a road-certified vehicle; synchronized concurrent throughput, latency, processor utilization, memory, frame drops, power and thermal behavior remain unmeasured.
Language: English
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7234795846
FWCI:
Cites: 34
Cited by:
Related to: 10