Get best of the week

Lidars at CES

image

Despite the fact that CES is an exhibition of consumer electronics, lidars and other components of robotic cars have become an important part of what it shows. At least 43 companies presented their lidars at CES, and some sources claim that there are currently around 150 different companies working in this area. If you look at the products of these companies, you can see that their designs are surprisingly rarely repeated - almost every device is made in its own way, and each company believes that their approach has a chance to bring them victory.

The main news of this CES was the participation of the lidar for robotic machines from Bosch, productivity growth, inexpensive lidars from various companies and several new participants presenting their design options.

Lidar companies seek to win in one or more of the following categories:

  1. The main success for the company is the use of their lidars for a full-fledged robotic car
  2. Other companies are striving to create lidars for the "autopilot", which helps the driver, although some of these systems (for example, in Tesla) do not use lidars.
  3. Cheaper devices with a shorter range that can provide visibility for navigation at low speeds and for observing objects close to the car that have disappeared from the field of view of the main lidar.
  4. Some companies rely on good business in other areas. Examples of such areas are low-speed robotics, smart city recognition, and security.

Lidar goals


In particular, in the first two categories there are many important methods of differentiation, and all these methods are aimed at achieving the following goals:

  1. Price: Taxis can afford expensive sensors, but for private cars they may not be available. And everyone wants to save, if there is such an opportunity.
  2. Range: For driving on a highway, a range of 200 meters or more is required. For lidars with a wavelength of 1550 nm, this is not a problem, and lidars with 905 nm still need to be further developed. Of particular interest are dark and difficult to detect objects, such as tires on asphalt.
  3. Reliability: everyone wants the device to continue to work and remain calibrated in a harsh automotive environment with lots of vibrations. Many early lidars had problems with this.
  4. : , , .
  5. : 360 . , . , .
  6. : , 10 , , , 20-30. .
  7. Special functions: measuring the speed of targets, preventing interference, capturing the entire scene in one flash - all these are additional features that can be pluses of the lidar.
  8. Size: The smaller dimensions and ease of integration into the car are the advantages of the lidar, although first users may prefer more obvious and understandable sensors like 360-degree devices mounted on the roof.
  9. Decoding: this includes the best hardware for decoding the return signal, and its connection with special software that helps interpret the generated point cloud.

Technology differences


To achieve all of the above goals, various lidars are needed. They can vary in wavelength, type of emitters and detectors, optics and how they direct rays (if they direct them) and how they process signals.

The laser wavelength is of great importance. Operating at 1550 nm (far infrared) allows you to safely use much more energy to increase the operating range. Also, this technology is quite expensive, since we cannot use inexpensive silicon microcircuits. The following relation works in the short infrared spectrum: the shorter the wave, the higher the efficiency of silicon microcircuits, but the greater the ambient sunlight. In these ranges it is difficult to achieve recognition of complex dark targets at a great distance, but many manufacturers claim that they can cope with this. Recognizing objects like retroreflectors (reflectors that can be seen on traffic signs and lane markings) is easy, but black cars, clothes, and tires are a different story.

Also, devices differ in their beam control methods. Some lidars, especially old and 360-degree models, simply rotate the entire lidar in a circle. You can see many devices with a small mirror that can vibrate (often in two dimensions) to direct the rays. Another popular approach is the use of microelectromechanical systems that can be embedded in the chip (they are often called fixed, but they have small moving components).

These solid state circuit approaches include phased array control (often found in radars) and frequency-based control (using a laser whose frequency can be quickly changed, and a prism that sets the direction of light change based on its frequency).

The new player in the market, Bajara, uses this method to control in one direction. The Strobe bought by Cruise is also rumored to use this method.

Flash lidars do not scan. Rather, they have a very large number of sensors (and possibly emitters) installed to capture the entire scene at once. It is very expensive, and it is also difficult to get a high viewing range, since a flash requires tremendous power. Thanks to cheap arrays of vertical radiation lasers, new companies like Sense Photonics are hoping to win with this approach, although at present they cannot provide a viewing range sufficient for driving on the highway.

These solid-state approaches are in demand, since it is believed that they will be the most reliable in harsh automotive conditions. Large moving parts are more difficult to maintain and calibrate. Nevertheless, as you can guess, each supplier now insists that its products are reliable and will not require frequent replacement or maintenance, and these are the requirements of automakers. Robot taxis that return to the fleet every night can only accept less reliable devices if it provides any other significant benefit.

Bosch


The most talked about newcomer to the show was Bosch. While they stated that they would launch their new lidar at CES, stepped aside and did not reveal any details except that their lidar will have a high range and its price will be appropriate for the driver assistance technology market. (As a rule, in the driver assistance market, sensors are needed that will cost significantly less than $ 1000, since no one wants to add thousands of dollars to the price of the car being sold. More expensive sensors are also suitable for the taxi market, since this price is distributed among passengers - a few cents per mile).

People pay attention to Bosch, as this company is one of the world's leading suppliers of top-level automotive components. No one has a better entry into the car manufacturers' market, and everyone knows this company well and trust it. This is a significant advantage over small and little-known startups. If Bosch creates a decent lidar, it is likely to outperform lidars from small companies. According to Bosch employees, they studied all the suppliers of lidars on the market, hoping to find the one they can purchase or enter into a partnership agreement. They found out that something was missing for all approaches, and decided to create a lidar on their own. They claim that their design is not identical to any of the already presented, even taking into account all the diversity. Their claim that they did not find a single company,which could be acquired may be due to the very high requirements that companies put forward in the field of robotic cars.

We will have to wait to find out more about the device that Bosch is building.

Focus on driver assistance


Another topic of the show was the so-called “winter of robot cars” - some companies said they were returning to work on driver assistance technologies. While the earliest enthusiasm concerned the use of lidars in fully robotic cars and taxis, most companies believe that more sales will come from the market for simpler and cheaper lidars, aimed at creating a competitor for Tesla autopilot (which does not use lidars but falls into some accidents that could have been prevented with their help).

Indeed, car makers will sell many driver assistance systems before they sell end-users fully functional autonomous cars. The latter is especially difficult, because users don’t like cars that work only in certain places, and they won’t bring the car back to the store every day for refinement (namely, how robots will work). A product such as Tesla autopilot has become a must-have feature for high-end cars (Tesla sells just those). Lidars can make such a product safer, and make it faster (if they fall in the price range). Not every OEM is ready to create such systems (or more advanced systems that will allow the driver to ignore the road while driving on the highway) without lidars.

Lower prices


Almost all companies predict that after the start of wide production, their lidars will cost less than $ 1,000. The average price of lidars from 1550 nm is from $ 500 to $ 1,000. At 905 nm - from 200 to 300. Chinese suppliers are promoting the low cost of their production, since they will be the first to set these prices. One of the most impressive players was Livox - they offered their lidars at prices ranging from $ 600 to $ 1,200.

Despite the fact that all companies need to promise low prices, many seem to be confident in their estimates, which means we can be sure that low-cost lidars from one of the suppliers will be available for several years - most predictions about robot machines just like that. This is important for the dispute between cameras and lidars, because one of the main arguments in favor of cameras was the high cost of lidars. Players using only cameras (like Tesla) rely on computer vision to work quite well. Players with lidars are betting on the cheapness of their sensors. Most likely, a bet on lidars will play. This whole discussion is far from simple, but cost is an important component. People like Elon Musk believe that lidars are not needed at any price and call them crutches. At the same time,computer vision has only one leg so far.

Speed ​​return


Some companies offer to provide measurements of the speed of moving targets using lidars. This is usually done using the Doppler effect, which can be calculated in devices that use continuous waves with frequency modulation - this is how most automotive radars work.

Last year, Blackmore, a company engaged in continuous frequency modulated radars, was acquired by Aurora. Several other independent companies (like Aeva and others) are promoting the same approach.

Knowing how fast the target is moving is very helpful. If you use a traditional lidar, then to determine the speed of the object you need to study a few frames. This can lead to a delay of 100 to 200 ms per frame, and then the same amount of processing. This can make a difference in critical situations (like an unexpected obstacle on the road).

Luminar, a leading player in the field of lidar at 1550 nm, has developed a different approach. They send several laser pulses to objects of interest every few milliseconds. If you do this accurately, then this is enough to determine the speed of the object. Using this technology (or using frequency-modulated continuous-wave radars) you can identify a standing object in a few milliseconds without any calculation, instead of spending 400 ms using computer vision or conventional lidars. Recent Tesla crashes involving collisions with cars parked in the left lane show how important this recognition technology is.

Who won?


Each company has its own story about how it will come to victory. Here are the factors that determine the winner:

  1. 1550 , .
  2. ( ) .
  3. , , .
  4. OEM- ( Bosch) .
  5. , . , , , .
  6. ( ) , 360 , .
  7. . , .

In the coming months there will be new topics for talking about lidars. Including such issues as the creation of a database containing information about all the major suppliers of lidars, and a deeper study of the confrontation between lidars and pure computer vision.

imageAbout the Author: Brad Templeton is a software engineer, an evangelist of robotic cars since 2007, worked on Google in his early years. Founder of ClariNet , honorary chairman of the Electronic Frontier Foundation and director of the Foresight Institute , founder of the faculty at Singularity University .


image

About ITELMA
- automotive . 2500 , 650 .

, , . ( 30, ), -, -, - (DSP-) .

, . , , , . , automotive. , , .

Read more useful articles:

More articles:


All Articles