Radar vs. Lidar: Best Value for Self-Driving Cars

Radar vs. lidar: What’s the best value for self-driving cars

Radar and lidar technologies are emerging as the major players when it comes to self-driving vehicles. While high-performance global positioning systems and cameras also play a role in the autonomous vehicle ecosystem, radar and lidar are the primary options. Ultimately, many vehicles currently on the street are designed to use a combination of all of these technologies, with one or two doing most of the heavy lifting. This is where the big decision comes into play – do you build a self-driving car around radar with other complementary technologies or use lidar with some amount of radar and other solutions?

In theory, this is basically a cost vs. performance decision, with lidar being more expensive than radar, but offering more detail for the underlying processing systems to use in making driving decisions. However, Karl Iagnemma, CEO of nuTonomy, a self-driving car startup, told the IEEE Spectrum that lidar offers a combination of size, cost and overall system complexity that makes it difficult to apply in its current form. This is beginning to change, however, making lidar a more viable option. With cost reductions taking place across the lidar sector, let’s take a deep dive into leading systems on both sides of the equation to look at the cost vs. performance balance in radar and lidar solutions.

Radar

Tesla is among the companies championing radar. According to the IEEE Spectrum, the automobile manufacturer is doubling down on the radar as CEO Elon Musk recently tweeted intentions to use radar by itself while using temporal smoothing. This would, theoretically, eliminate the need to have a camera working alongside the radar system while also providing some degree of dimensional analysis, which is what makes lidar stand out. If this strategy pays off, it could offer a viable, cost-efficient alternative to lidar.

Details on the makeup of Tesla’s radar system are scant as the technology is proprietary and most self-driving car makers developing their own systems keep details close to their vests. However, a look at the S32R27: S32R Radar Microcontroller from NXP Semiconductors provides a brief glimpse at the cost and system expectations of a radar system for a self-driving car. The S32R27 unit is a microcontroller functioning on a 32-bit power architecture and provides robust integration for automotive radar. This allows it to merge radar signal processing with microcontroller functions with a multi-core architecture.
Self-driving car radar units featuring the S32R27 can use three major radar interfaces – MIPI-CSI2 with four data lanes, DAC at 10 MSPs or a ΣΔ-ADC setup in a 4×12-bit, 10 MSPs configuration. The system is certified for ISO26262 SEooC safety standards and can function at temperatures ranging from -40 degrees to 150 degrees Celsius. In terms of compatibility and connectivity, the system works with Zipwire, 2x SAR-ADC, 2x SPI, 2x I2C, 3x FlexCAN (including 2x CAN-FD), FlexRay, LINFlexD and Ethernet interfaces.

All of this comes together in a $375 hardware-only package, but that represents the base unit price, which can vary depending on the seller and on how many units are being purchased.

Lidar

Velodyne is among the industry leaders in the lidar space, and it recently released a new all-in-one LiDAR “Puck” product that aims for a $7,999 price tag. The puck simplifies many traditional lidar functions by using 16 lasers to map the world around the vehicle. Velodyne’s previous industry-leading systems used as many as 64 lasers, which created subsequent cooling and equipment demands that drove costs upwards. Typically, a 64-channel lidar unit will cost approximately $85,000, with 32-channel systems ranging from $30,000 to $40,000. Wolfgang Juchmann, head of sales and marketing at Velodyne, explained that these ultra-high-performance lidar units made sense for research purposes, but the lower cost 16-channel solution is a much better fit for mass production.

The Velodyne LiDAR Puck provides real-time 3D environmental processing using approximately 8 Watts of power and in an 830-gram footprint. It offers 360-degree horizontal visibility with a 30-degree vertical field of view and a range-extending out to 100 meters. It can operate in temperatures ranging from -10 degrees to 60 degrees Celsius. In terms of networking, it supports a 100 Mbps Ethernet connection.

Velodyne isn’t the only lidar company out there, but it has come to set the standards in the industry, including being used in Google’s prominent self-driving car project, and it sets a benchmark for the rest of the sector.

Radar vs. lidar

In its simplest form, this debate comes down to a significant functional difference compared to a chasm in capital expenses required to invest in the technology. Radar is considerably less expensive than lidar when considered on a unit-by-unit basis, and the price difference changes exponentially when considering the need to build out an entire sensor ecosystem. For example, a report from Quartz pointed out that a smaller lidar device that would be priced at approximately $8,000 would typically be deployed across four locations in the car, bringing the total cost of the lidar set up to $32,000. An entire radar ecosystem that would accompany lidar, on the other hand, would cost $10,000.

A more robust radar system may cost more than the $10,000 mentioned, but Quartz was also assuming that a lidar system would require Velodyne’s $85,000 HDL-64E system combined with the four smaller setups. When combined with other peripherals, including radar, cameras, GPS and computing systems, not to mention the actual car, you’re looking at a total cost of $300,000. This changes dramatically if the Velodyne Puck emerges as a standalone lidar solution or if Tesla’s radar option that uses temporal smoothing can get the job done.

Regardless of how all of this pans out, there is still a divide in cost vs. performance, and choosing which is the best fit may depend on your project. If you want to introduce autonomous vehicle functionality into automotive models (or robots), but still have the system rely on a human driver paying attention, the advanced radar may be an option at a market-ready price. If you are trying to push for fully autonomous vehicles, however, you may need to take a look at the emerging lidar systems that offer a better price. This debate is key to the industry’s evolution, but alternatives are already on the horizon.

Emerging options

Radar and lidar aren’t the only self-driving car vision options out there. Besides cameras and GPS units – which typically need some combination of radar and lidar – emerging solutions are rising to take the market in new directions. In particular, startup Oryx Vision is developing a solution that receives light as waves instead of as particles. This allows for much greater depth resolution than lidar with the high range of the radar. All of this happens in a solid-state system, meaning it won’t have moving parts that lead to mechanical failure over time.

The Oryx system is still extremely new and it isn’t much available in terms of details, but it is a reminder that the self-driving car sector is still in the rapid development phase. Because of this, project leaders may want to continually stay abreast of developments as they emerge to ensure they find the best price-to-performance ratio. Choosing between radar and lidar requires some project-specific decision-making, and may come down to the type of functionality you are looking for. All told, however, lidar offers some advantages as ongoing evolution in the sector could bring the costs down while continuing to refine the performance advantage it has over the radar.