As lidar technology becomes the industry standard in powering autonomous vehicles, a young startup called Altos Radar is stepping up to challenge the light-based remote sensing technology with 4D millimeter wave radar.
Based in California and founded in January, Altos Radar recently raised its first round of funding of $3.5 million from ZhenFund, Monad Ventures, and Yifan Li. The participation of Li, the founder and CEO of Hesai, a major lidar maker that racked up $190 million in an IPO in February, seems surprising at first given the battle between radar and lidar to win the AV clients. At a closer look, however, the investment indicates an interesting new development in the arena of automotive perception.
Lidar, which uses light to measure distances between objects, is currently considered more robust than radar in providing high-resolution, three-dimensional mapping. But there’s a trade-off: High-end lidar sensors can easily cost tens of thousands of dollars, though Chinese manufacturers like DJI-affiliated Livox and RoboSense have brought their costs down significantly.
Li Niu, co-founder and CEO of Altos Radar, is convinced that millimeter wave radar is advancing at a pace that makes it a strong substitute for lidar in advanced driver assistance systems (ADAS) or even autonomous driving.
“Lidar only came to the fore as autonomous driving emerged. In the early stage of development, companies worked to make the sensors as powerful as possible at all costs. But as we progress into the second half of the competition, the focus shifts towards generating tangible value,” suggested Niu.
A battle: radar, lidar and camera
Altos’s automotive radar, according to Niu, is superior to lidar on several fronts. For one, it has a 350-meter sensing range, which is longer than most lidar products and comes in handy for highway driving. It works in most weather conditions, given radar’s ability to penetrate objects. Importantly, Altos is able to measure instant and accurate speed at 0.2m/s, which, according to Niu, is “important to predicting a vehicle’s trajectory.” In addition, its sensors can distinguish objects that are as close as 0.31 meter apart.
Some of these are common qualities of radar and aren’t necessarily unique to Altos. “Almost all of them can be improved,” said Niu, but they are “design choices” and the improvement of some could lead to compromised capabilities in others.
Given all these benefits of radar, including its affordability at a fraction of lidar’s costs, why hasn’t it been widely adopted in AVs? Niu pointed out that the 77GHz band that automotive radars use is a new standard that’s only been available since 2017. Due to the low resolution of incumbent radars, they’ve only played a “supporting role” in providing velocity information.
Altos pledged to be different. In a pre-recorded demo (below), Niu showed that the startup’s radar is able to generate “point cloud” data, which is real-time, high-resolution representations of moving objects, a capability for which lidar is known. This, according to the founder, is Altos’s real differentiator.
Major Tier-1 manufacturers have also been working on high-resolution radar. But most of their solutions, Niu argued, aren’t production-ready, for they tend to use field programmable gate arrays (FPGA) as processors, which require complex programming and a great amount of computing power.
Altos, on the other hand, uses the application-specific integrated circuit, or ASIC, which is optimized for a specific purpose resulting in lower costs and power consumption. This is done through “compute optimization,” according to Niu, meaning the sensors can get much more computing power out of the same chip. Specifically, the startup claims to be able to achieve 80x the performance as an ASIC’s reference radar design.
How does Altos compare to cameras, which Elon Musk famously declared are the future of AV perception? Radar consumes much less computing power, as it “only streams useful information” and “offers the extra perk of measuring distance and speed,” said Niu. In 2021, Musk ordered radars to be removed from Tesla vehicles, but that was due to the frustration of their poor quality at the time and he still believed that “higher definition radars” would improve Autopilot and Full Self-Driving, according to his interview with Electrek in February.
Software vs. hardware
Niu argued that his startup’s other moat is his team. At Apple, he was one of the first 150 employees to work on the giant’s autonomous system hardware from scratch. Later at Pony.ai, a Toyota-backed robotaxi upstart with offices in China and the U.S., he led the in-house radar and camera team.
Michael Wu, another co-founder, learned from his previous experience as a mobile platform engineer at Mozilla where he specialized in optimizing browser performance on end devices. Also a veteran of Pony.ai, his primary role now is to ensure that the solution provided by Altos is free of software delays in vehicles.
These experiences equip Altos with the know-how to work at the intersection between software and hardware, Niu said.
“When it comes to dealing with software and hardware, there are historically two camps. One is represented by Apple, which works on hardware and software simultaneously; the other is Android, which focuses on software and delegates hardware to OEMs. Tesla is the Apple of the new era,” said Niu. “It’s hard to say which camp is better, but I personally think the best-in-class companies work on both.”
“Hardware production is actually the easy part. Our competitive advantage is our R&D and software design,” he added.
Altos’s radar products are “ready to ship” and the startup is in early discussions with dozens of customers ranging from original equipment manufacturers and self-driving companies to universities and port facilities in both the U.S. and China.
Despite a well-rounded team and a seemingly competitive product, Altos faces several challenges ahead. Venturing into a deeply established industry with a complex and extensive supply chain is no small feat to start with. The startup’s success also hinges on multiple stakeholders along the value chain.
“Our customers, precisely speaking, are not OEMs themselves, but rather the autonomous vehicle divisions within these OEMs. Our main challenge will be whether these AV teams can effectively generate value for their end users and effectively make use of millimeter wave radar, which is a field that few are familiar with,” he noted.