SENSORS In addition to advanced sensing, lidar has overcome some of the hurdles it faced in the past, including costs and durability. There are now several lidar systems that are being integrated into L3 autonomous vehicles. Lidar manufacturers are also now addressing different market segments. “There are two types of lidar makers,” says Jeremy Cohen, founder and CEO at Think Autonomous. “There are those who sell to auto manufacturers that bundle the lidar with the car, so they have to get their sensor to be very cost-effective.” Examples include Innoviz Technologies, which is selling lidars at US$500 to US$1,000. The other notable market segment is robotaxi providers that use the sensors in their automated ride-hailing services. These companies can afford higher-priced lidars, such as those supplied by Ouster. “They can go to US$10,000 to US$15,000 because they’re not going to sell the car with the lidar – they’re going to sell a service that will compensate for the costs,” Cohen explains. Robotaxi services have hit some bumps in recent months, but if they manage to iron out the kinks and reach mass scale, there will certainly be a growing market for higher-end AV lidars. “WE’RE REACHING A STAGE WHERE YOU CAN BUILD A POINT CLOUD WITH A RADAR THAT CAN DIRECTLY COMPETE WITH THE LIDAR” Jeremy Cohen, founder and CEO, Think Autonomous Radars are making a comeback Radars, which some experts believed were excessive and unnecessary, are becoming increasingly useful for different tasks. According to the SAE report, radar is now used for a growing number of automotive applications, including some that were traditionally covered by other sensors, such as ultrasound devices. “This is because radar has reached a price point where it becomes cost-competitive while offering more range, higher accuracy and additional functionality, leading it to become a universal sensing concept for many kinds of automotive systems,” the report states. “Radars are interesting,” Cohen says. “They are no longer the super-noisy sensors that you could almost remove from the car. We know it’s useful in a way, but it’s not a standalone sensor. And I’ve seen a lot of companies doing great work on radars and imaging radars too.” Classic 3D radars measure the distance, azimuth and velocity of objects; 4D radars, sometimes called ‘imaging radars’, use additional vertical beams to detect the height or vertical position of an object so that it can be classified more precisely. And more advanced radar systems can generate point clouds that compete with lidars and can provide more precise object detection. “This could very well complement the camera LIDARS IN EARLY solutions or even replace some lidar solutions. Now AUTONOMOUS we have a dense point cloud, the noise can be filtered, VEHICLES COST and we can use deep learning on top of that and really improve the radar output,” Cohen says. “We’re reaching AS MUCH AS a stage where you can build a point cloud with a radar US$75,000 that can directly compete with the lidar.” Recent years have seen tremendous progress in AI and computer vision systems. Deep learning algorithms can segment images and detect objects with increasing precision and efficiency under different conditions. But manufacturers of ADAS and AV Sensor redundancy still rules Mobileye’s True Redundancy uses a lidar and radar perception system 28 ADAS & Autonomous Vehicle International April 2024