By ATN Editorial Team
Author: LeddarTech CTO, Pierre Olivier
At Level 3, the vehicle can handle all aspects of driving within certain conditions or environments. This includes tasks like steering, accelerating, braking, and monitoring the surroundings. Unlike Level 2 (Partial Automation), where the human driver must continuously supervise the system, Level 3 allows the driver to disengage from the driving task entirely during specified conditions.
One of the primary challenges is ensuring the safety and reliability of Level 3 autonomous systems. These systems must be capable of handling complex driving scenarios and making split-second decisions to avoid accidents, especially when transitioning control back to the driver. Additionally, the accuracy of these systems in perceiving and interpreting their environment is critical. This includes the ability to detect and respond to obstacles, lane markings, and other vehicles accurately, even in challenging conditions, such as poor weather or low light.
From a commercial standpoint, developing and deploying Level 3 automation must be cost-effective and scalable. The technology needs to be economically viable for mass production while maintaining high performance standards. Moreover, these systems must comply with varying regulatory frameworks across different regions. They must meet stringent safety standards and gain regulatory approval to be deployed on public roads.
At LeddarTech, we believe that many of these challenges can be addressed through better perception technologies, particularly with our LeddarVision low-level fusion software. LeddarVision leverages advanced AI and low-level fusion technologies to enhance data interpretation, reduce latency, and improve accuracy and reliability. This paves the way for safer, more efficient autonomous driving systems, making Level 3 automation a viable reality.
One of the biggest challenges is achieving the right level of performance under difficult weather and lighting conditions. It is a fact that a disproportionately high number of accidents occur under such conditions, for instance at night. This is where Automated Driving (AD) systems can have the largest safety benefit. For example, driving in winter conditions can be very challenging compared to driving on clean, dry roads because the driver does not usually have the benefit of lane markings. These challenges also occur during heavy rain and in poor lighting. That’s where our AI-based low-level sensor fusion and perception technology takes over by extracting and combining data from the different sensors rather than relying on isolated sensor inputs, which is essential as the fusion process compensates for sensors that may be compromised due to various factors. By integrating raw data from multiple sensors, we are creating a single cohesive model of the surroundings of the vehicle, improving accuracy, safety and reliability for the driver.
To address the commercial headwinds in large-scale deployment of Level 3 systems, one must look at the system cost. Another benefit of LeddarVision is the reduction in the number of sensors required. We have demonstrated solutions where, instead of using 11 cameras and 5 radars, we can achieve equivalent or better performance with just 5 cameras and 5 radars. This results in a significant cost-benefit for auto manufacturers, as they not only reduce the number of cameras and sensors per vehicle but also save on associated components like wiring and computing resources. Reducing the number of sensors is key to meeting the cost and price points expected by the industry.
Sensor fusion and perception accuracy are critical, even under optimal conditions. For instance, lane keep assist systems can often be fooled by imperfect lane markings or metal guard rails. In a Level 2 system, the driver can immediately react and correct, but in a Level 3 system, the accuracy of lane estimation needs to be high enough to prevent such errors. Optimising processing power is also necessary, given that computing capacity is constrained by the current state of semiconductor and SoC technology. Based on Moore’s law it is predicted that it will take 10-15 years before we can envision processing approaching the human brain embedded in a car, so very tight optimisation of computing resources will be essential to meet market expectations.
Finally, the regulatory frameworks need to allow for Level 3 deployment. This is where standardisation of not only performance but also the sensor fusion and perception software that is implemented, such as low-level fusion, provides the confidence for regulators to approve Level 3 automation globally. By addressing these technical and commercial challenges through advanced perception technologies, we believe that consumers will embrace the enhanced safety and greater reliability that Level 3 automation provides.
For technology news from LeddarTech, click here.
