DVN: What is Mobileye’s view on what kind of hardware and software will be needed for the FMVSS 127 AEB regulation?
Shimon Solodkin: Mobileye’s current camera-based systems demonstrate strong performance in nighttime AEB scenarios, including with only a single front-facing camera. For FMVSS 127, achieving compliance largely depends on the overall system design, including headlamp optimisation on the automaker side. Our current assumption is that a single camera shows very good performance against FMVSS performance metrics. Mobileye’s camera-based solution leverages advanced image processing and ‘AI’ algorithms achieve high accuracy for pedestrian and vehicle detection, even in challenging low-light conditions, without requiring additional sensors. Additional sensors beyond cameras would add cost and complexity barriers that could limit widespread adoption.
DVN: Are LWIR thermal cameras a better solution?
S.S.: Mobileye believes optimizing the capabilities of forward-facing cameras, combined with robust ‘AI’ processing, offers a more cost-effective and scalable solution for meeting regulatory requirements.
DVN: How about HD radar?
S.S.: Radar’s resolution for detecting smaller objects or differentiating pedestrians from other obstacles is less precise than camera-based systems on their own. Mobileye integrates radar with cameras in some configurations to enhance redundancy and performance, especially for higher levels of autonomy.
DVN: Lidar is still expensive; how does HD radar compare?
S.S.: The imaging radar solution we are preparing to bring to market demonstrates the potential to reduce the need for surround lidar in advanced systems, offering cost advantages to automakers without sacrificing performance. However, we believe Mobileye’s camera-based solutions will be capable of meeting regulatory requirements without the need for the added cost and complexity of imaging radar or lidar. Imaging radar and lidar become more relevant as automakers deploy more advanced automated systems, such as eyes-off highway driving.
DVN: What should the ‘AI’ component of an AEB system look like?
S.S.: ‘AI’ models for driver assistance applications are most efficiently deployed on purpose-built hardware optimized for ‘AI’-specific workloads. Mobileye’s EyeQ family of SoCs are designed from the ground up to deliver high-performance, low-power processing tailored for ADAS and autonomous capabilities. Our processors can handle the demands of multi-sensor fusion, including both cameras and radar, with enough computational power to enable real-time object detection and classification.
As an example for ‘AI’ integration, the EyeQ6L harnesses a deep learning dynamic neural network (DNN). When combining the additional computational resources EyeQ6L offers with new ‘AI’ integrations, this unlocks powerful capabilities, like Neural Network Semantic Segmentation (NSS), our pixel segmentation DNN, and pairs it with advanced classifiers, which enables “pathways”. Pathways are in reference to a novel algorithm that introduces center paths for all lanes in the image simultaneously. There are also multiple new sensing products, such as advanced, any-object detection and enriched semantic information of vulnerable road users and vehicles alike.
DVN: What is the optimal sensor set for L2+ including AEB?
S.S.: By 2027, we anticipate L2+ capabilities, such as eyes-on/hands-off highway autopilot, will be offered as a premium option by most global automakers. Mobileye’s scalable architecture supports both cost-effective base configurations and enhanced systems with additional sensors for L2+ to L3 functionality. For these requirements, our Surround ADAS solution, using front, rear, and parking cameras linked through a single EyeQ6H, provides robust safety and driver-assist features with a simpler sensor configuration and lighter computing load, making it a cost-effective option for mass market vehicles. At the premium end, Mobileye SuperVision™ delivers robust, hands-free L2+ functionality in defined domains through 11 cameras for 360-degree coverage, as well as an HD Map and Mobileye’s RSS driving policy model, using two EyeQ6H—building on the economies of scale enabled through Surround ADAS.
DVN: How does camera detection of objects differ from human perception?
S.S.: “Perception” is the key term—humans perceive objects, the environment and key driving tasks in many ways that surpass even the most advanced camera systems. However, cameras have the advantage of never being distracted or fatigued from sensing what’s around the vehicle. Our long history with camera-based safety systems proves that technology working together with humans saves lives and reduces crashes.
DVN: Are there requirements on low beam performance for good AEB performance?
S.S.: Yes, there are minimum requirements already defined in headlamp regulations. Mobileye works with automakers to align headlamp design and sensor performance to meet these standards effectively.
DVN: What would an optimum low beam look like as far as sensors are concerned?
S.S.: FMVSS 108 defines the maximum allowable illumination for low beam performance, which serves as a key benchmark for AEB nighttime performance. Mobileye is actively collaborating with automakers to strike the right balance between optimizing illumination for sensor performance and ensuring compliance with FMVSS 108.
DVN: In a camera system, what is the time span for object detection?
S.S.: Our front-camera system is capable of detecting objects at ranges of hundreds of metres during daytime conditions. In low-illumination scenarios, such as those specified under FMVSS 127, detection range is more constrained due to the more limited visibility provided by low beams. However, the nighttime breaking scenarios outlined do not require extremely long detection distances for AEB to be effective. Detecting an object as close as 40 metres is sufficient to ensure safe braking under such conditions.
DVN: What do you think about low beam for distant pedestrian detection under streetlights?
S.S.: Detection with low beams under street lighting has been a standard part of NCAP testing for several years. We don’t see this as a significant challenge, as current systems on the market have consistently demonstrated strong performance at speeds up to 60 km/h as required by NCAP. We work frequently with automakers to ensure low-beam headlights provide enough light for robust camera sensing in these situations at a variety of speeds.
