Best Practice

Improving the safety of autonomous driving through the detection of traffic lights

By harnessing the power of machine learning, Intel Habana Gaudi processors, and AWS, Concept Reply created an advanced AI-based solution that allows autonomous vehicles to accurately detect traffic lights, improving road safety in urban areas.

Making roads safer for all:
co-producing excellence with Intel

The traffic light detection solution offers a practical and effective approach to improving autonomous vehicle navigation and was developed in partnership with Intel. Intel is a leading force in the industry, driving innovation in hardware, software, and service technology. With a focus on integrating intelligence and AI into every computing device, from the cloud to the edge, they're harnessing the power of data.

The solution optimizes Intel processors for machine learning training, by introducing an innovative method that surpasses GPUs in terms of efficiency. It integrates the Habana Gaudi AI accelerator to offer effective and profitable AI training. Advanced machine learning features are implemented using YOLOX for object detection and PyTorch for deep learning. Leveraging AWS hosting, it ensures scalability. This synergy facilitates the accurate detection of traffic lights in real time, which is essential for the safety and navigation of autonomous vehicles in dynamic urban traffic conditions.

Leveraging simulation to refine models and data

To refine autonomous driving models, we rely on CARLA, an open source simulation environment. CARLA not only facilitates rigorous model testing, but also accumulates valuable data to further improve models. CARLA offers us numerous advantages, allowing us to test and refine our autonomous driving models while collecting essential data to refine them. CARLA simulates complex automotive sensors, including GPS, LIDAR, and accelerometers. In addition, it offers tailor-made driving scenarios, allowing simulations that replicate challenges such as driving at night or various weather conditions. This innovative approach allows us to refine our algorithms and improve the accuracy of traffic light detection.

Ensuring the reliability and accuracy of CARLA involves extensive training in machine learning. Faced with a shortage of GPUs, Reply turned to Intel for an innovative solution. Intel, a leader in hardware and AI, collaborated with Reply to create a powerful and scalable platform optimized for cloud instances. This solution adapts dynamically to different vehicle numbers, giving priority to resource efficiency and sustainable computing.


See it in action

Key strengths

Reliable detection of traffic lights

The solution uses machine learning based on YOLOX and PyTorch to provide accurate and reliable detection of traffic lights, a component that is critical to the safety of autonomous vehicles.

High performance processing

Equipped with Intel processors and the Habana Gaudi AI accelerator, the solution offers high-performance processing without compromising its price, making it an exceptional choice.

Scalability and flexibility

Integration with Amazon Web Services (AWS) ensures the scalability and scalability needed to meet different traffic scenarios. Your autonomous vehicles will always be ready.


Effective training

Thanks to the Habana Gaudi AI accelerator, the solution offers effective AI training on processors, making it a prudent choice for organizations looking to optimize their resources.


Real-time detection

The solution excels in the rapid and accurate recognition of traffic lights in real time, an essential feature for autonomous vehicles facing the dynamic challenges of urban traffic.



Our strong point is to find the ideal balance between high-performance computing and accessibility, thanks to the effective design of the Habana Gaudi artificial intelligence accelerator.


Find out how Reply contributes to success in the automotive world

Concept Reply

Concept Reply is an IoT software developer specializing in the research, development and validation of innovative solutions and supports its customers in the automotive, manufacturing, smart infrastructure and other sectors in all matters related to the Internet of Things (IoT) and cloud computing. The objective is to offer complete solutions throughout the value chain: from the definition of an IoT strategy to the implementation of a concrete solution, including testing and quality assurance.