The realization of autonomous driving is inseparable from LiDAR

Thanks to the auto industry's constant tinkering with new words, we now know that driverless cars are "coming soon." Some manufacturers predict that self-driving cars will enter the commercial market as early as 2021. However, the new technology will be very expensive when it arrives, and most ordinary people will not be able to enjoy it for a while. Let's follow the automotive electronics editor to learn about the relevant content.

LIDAR, the imaging system that allows self-driving car software to see the road, is ridiculously expensive. High-end systems can cost close to $100,000, while low-end systems can cost less than $10,000. Emerging technology is never cheap, but historically, manufacturers have found ways to make components cheaper. However, the time left for self-driving cars is too short, which means that if manufacturers want to launch self-driving cars, they must either accept high costs or delay production.

 The realization of autonomous driving is inseparable from LiDAR

Yes, automakers can choose cheap hardware to keep costs down, but it's not worth the risk. Why? Because car autonomy places a huge responsibility on manufacturers, and even mid-range lidar systems are a gamble in terms of safety. According to MIT Technology Review, low-end lidar systems are useless when cars are driving at highway speeds.

MIT used data from two Velodyne systems as an example, one is the HDL-64E, which costs $80,000, and the other is the Puck, which costs only $8,000. The HDL-64E uses 64 laser beams to draw images of the surrounding environment with high clarity, covering up to 120 meters, while the Puck has only 16 light spots, and the image will soon be distorted, with a maximum coverage distance of about 100 meters. If the car is driving at a low speed, both systems are good, but if it is driving at normal driving speed, the Puck will not work.

The problem is that even the HDL-64E system is only just meeting the minimum requirements for highway driving. When the car reaches a speed of 70 miles per hour, the car moves forward 31.3 meters per second. If an emergency occurs, the car needs a distance of up to 60 meters to come to a complete stop. Every meter beyond that gives the computer more time to make decisions, and the longer distance is absolutely necessary for the car. Ideally, the car's lidar system only provides an effective image of 200 meters away, which is safe when driving at highway speeds.

There is a technology that can solve this problem, and that is a solid-state system, which is much cheaper than traditional lidar systems. Quanergy claims that it has made a $250 component that is used in the so-called S3 car. Unfortunately, the system has low fidelity and is only useful when the car is driving at ultra-slow speeds (under 10 miles per hour), otherwise it is useless. Velodyne is also developing cheap solid-state components, but it promises that solid-state systems are not yet a substitute for 360-degree laser systems.

Austin Russell, CEO of Luminar, explained that the sensors they use try to avoid solid-state hardware. Why? Mainly because although the laser system is much more expensive, its image is also much better. To make driving safer, the clearer the image, the better. He also said: "If you only have a few points about an object, no matter how powerful your machine learning technology is, it will be useless because you don't know what it is. If you can only see something 30 meters away, when the car is driving at highway speed, it is only one second away."

Graeme Smith, director of the Oxford University autonomous driving project, believes that the lidar industry may seek a balance between data quality and price, which will lead to differentiation and make the time for high-speed autonomous vehicles to be put on the road different. He said: "The price of low-speed versions may drop faster than that of high-speed autonomous vehicles. If you want the laser to cover more than 250 meters, you need to use a better laser. If you only drive in a low-speed environment, you can only cover 15 meters, so you can use a cheaper sensor."

Still, installing low-quality lidar on cars will increase costs. Industry researchers and automakers believe that self-driving hardware will add $8,000 to $10,000 to the final price of the car, but the actual cost may be higher. When you add in all the hardware in the existing test platform, the final price is astronomical. Tesla says it wants to get the radar system down to $8,000, which looks beautiful, but it will also install thousands of dollars of computer equipment, cameras, and inertial measurement units (which can come in handy when GPS is not available). Many people worry that the images drawn by Tesla's radar are not clear enough.

In addition, if companies use high-definition lidar systems, they still have to switch to radar or shut down the system in foggy weather. In a recent interview with Axios, Luminar's Russell said that if manufacturers want to retail self-driving cars, they must do their best to ensure safety, reduce the "critical failure rate" and ensure that the image is extremely clear. Russell said that the self-driving industry has placed too much emphasis on reducing costs, and now developers need to focus on enhancing the safety of the technology to make it unbreakable.

Russell believes it will probably cost between $300,000 and $400,000 to perfect the technology, a price fleet owners are willing to pay because ride-sharing is a profitable business.

The above is an introduction to the fact that the realization of autonomous driving in automotive electronics is inseparable from lidar. If you want to know more related information, please pay more attention to eeworld. eeworld Electronic Engineering will provide you with more complete, detailed and updated information.