What went down at CES 2023 for RoboSense?

     RoboSense, the leading LiDAR company in China based in Shenzhen, debuts at CES 2023 their first true solid-state automotive-grade LiDAR: the RS-LiDAR-E1. The next generations of the M-Series Long-Range MEMS automotive-grade LiDAR were also shown.



     The E1 packs a punch aimed at current ADAS and AV(Autonomous Vehicle) peripheral-view sensors. (Looking at you, radar). Its purpose is to replace the blind-spot radar in ADAS cars and to eliminate blind spots for AVs. It will do this by being offered at a comparable price to the radar for automakers, the price being enabled by RoboSense's all-in-house chipsets, development, and manufacturing.

     The E1 is much more than a replacement for radar, though. A blind-spot radar can reliably tell you that "there is something in your blind spot area." In contrast, the E1 will tell you exactly where the object is, the size, shape, distance, reflectivity, velocity, and whether it's a car, truck, cyclist, or pedestrian, all this at 25 times per second, out to 50m or 30m @10% reflectivity-the industry standard designation for LiDAR range. The ultra-high refresh rate and more extended range are crucial to making safe turns and corners, as demonstrated by the following video graphics:


     As displayed, the E1 at 25 Hz(Right) can sense the biker coming from the right at 8 different positions on his trajectory as he approaches the crosswalk. The 10 Hz simulation(left) only saw the biker at 3 points and nearly sent him to Jesus before stopping.

      Also, the E1 allows the car to see fast traffic from the sides in an unprotected left turn, avoiding a fatal crash, where a blind-spot radar would be essentially blind.

     Before the M1 and E1 sensors, LiDAR was too expensive for the average consumer car because you would need 360 coverage with LiDAR, as shown in the image below. Without the E1, the car would need at least 3 M1 sensors or a combination of M1 with RS-BPearl mechanical hemispherical LiDAR, which is not automotive grade due to the internal mechanisms. Now with the E1, a seamless 360 LiDAR system will be affordable, aesthetically pleasing--if not invisible (Try to find the M1 on the 2022 Lucid Air, pictured lower in the article), and higher performance due to the 150% frame rate increase. The most exciting aspect of the E1 will be its ability to be adopted by base trim-level cars enabling Level 3 self-driving at lower speeds for all cars equipped with the E1 (A forward-looking long-range LiDAR is necessary for higher-speed L3 Self Driving).


     The long-range of the M1 is overkill in a 360-degree deployment for Level 3 Autonomous Driving, and the 10Hz frame rate is less ideal than the E1's 25Hz. However, that was the only automotive-grade option available at the time. Now with the E1, the full 360-degree coverage will be affordable, with options for the M1 mounted facing forward only as required for high-speed L2+/L3 autonomous driving. 360-degree LiDAR coverage is necessary because a blind spot will cause Perception Software to misjudge overtaking or overtaken vehicles to the sides of the car as two objects as they exit one LiDAR's horizontal FOV, and enter another LiDAR's view as depicted below:

     This affordable L3 solution will be possible because The E1, and LiDAR, in general, require less processing and calibration while providing a significantly more accurate scan of the environment than radar and Cameras. LiDAR's better accuracy comes from its factual data about the road detected as actual points in space (Point-Cloud), in the millions of points per second. Then through mature perception software, the objects are recognized and classified, with a velocity and a box of appropriate size, like a "hit box" to those familiar with FPS video games. Oh, and it can track hundreds of these objects all at once. This is opposed to cameras and radar working together to roughly guess where an object is, at the expense of complicated and unreliable processing by sensor fusion and inference.

     Developing the camera/radar-based perception software is much more complicated and expensive than LiDAR-based software due to the lower-quality sensor output. Cameras give 2D images, but range detection has to be inferred from various factors like object size, velocity, and different perspectives. The radar is also necessary to provide accurate ranging information. These must then be fused together to provide usable data to try and perceive the environment. Raw LiDAR output can be visualized and easily perceived by human eyes as a 3D image/video, and it is easy for the software to see scanned objects and roadways as a live 3D scan of free space and obstacles. The obstacles are also easy for the software to recognize and classify as various sizes of vehicle, cyclists, pedestrians, traffics cones, curbs, etc. Even if it doesn't recognize an object, it will still "see" everything as it sees actual points in space.   

     Additionally, these sensors must be calibrated and positioned precisely, limiting their availability to high-end trim levels with costly maintenance. LiDAR is immune to these issues as it is independent of different sensors to provide accurate perception; the point cloud can be fused seamlessly in real time before processing. LiDAR's FOVs can be overlapped to eliminate needing fine positioning. The result is a seamless 360-degree point cloud of the environment around the car-depending on the positioning of the sensors. This is not to say that RADAR and cameras will be eliminated entirely; LiDAR will only enhance existing perception methods. Nothing available can beat RADAR in adverse weather for reliability and cost, and Cameras are inexpensive devices capable of recording necessary 2D high-resolution, high-quality images and video. All are important, and each has its advantages and will continue to be fused together to ensure the safest detection possible. For reference, pun intended, RoboSense's own RS-Reference Ground Truth System uses 3 LiDAR, 6 Cameras, 2 RADAR, and RTK with IMU (100x more accurate than GPS alone) to generate the highest quality environmental perception data available to verify in-development ADAS and AVs.


     The M1 sensor was the first mass-produced automotive-grade MEMS LiDAR. It is the most produced and utilized sensor in China.

     RoboSense demonstrated the M-Series Sensors at the booth to the enjoyment of visitors, and several partners of RoboSense displayed the sensor.

     The M series is based on RoboSense in-house developed 2D MEMS smart scanner chips and achieved the technological breakthrough of advancing from 1D mechanical scanning to 2D chip-based scanning. It features high performance, low cost, high reliability, and high scalability, leading LiDAR products into the era of large-scale mass production. The M series is the only LiDAR product that can continuously improve performance. Seamless iteration and upgrading can be achieved under the same size, installation specification, connector, and protocol conditions, which cater to the different needs of car company customers for dynamic upgrading of smart perception solutions. 

     In particular, RS-LiDAR-M1 (M1 for short), the first M-series product, has won the Innovation Awards at CES 2019 and CES 2020 for two consecutive times. It is the only second-generation smart solid-state LiDAR worldwide to achieve automotive-grade series production and delivery. It has won a series of production contracts from 10+ famous automotive brands, including BYD, SAIC IM, FAW Hongqi, Chery, GWM, XPENG, Zeekr Intelligent Technology, Lotus Cars, GAC AION, WM Motor, and Lucid. Among them, series production projects cover over 50 vehicle models, including sedans, coupes, supercars, SUVs, and heavy-duty trucks, and the forecasted order quantity is expected to exceed 10 million units.