Spartan radar test van
Spartan’s test van at work: visualization of our Ago software intelligently applying Super Resolution to objects in a scene.

Many sensor companies talk about range, resolution and cost. At Spartan, we focus on how sensors perform in critical on-road scenarios and what our software can do to make these scenarios safer. That’s why one of Spartan’s primary goals in the summer of 2021 was to acquire and build a research vehicle for software testing, demonstration, and to have a rolling lab that can find complicated driving scenes where they happen: on the road.

I’ve been fortunate to help customers design these types of research vehicles in the past, back when I was working for AutonomouStuff. It’s really exciting work. So, I took the lead on this project—from vehicle selection and design to the acquisition of computers, racks, power converters, GNSS/INS, and other sensors.

Today, we have that branded research and development vehicle. It not only supports customer demos but enables us to further develop and validate our software.

Uncovering radar tech limitations—Record these situations in the wild, then develop specific solutions.

Since building our van, we’ve asked hundreds of potential customers to describe the shortcomings they experience with radar in their ADAS and automated platforms.

Here are some of the issues we heard:
● Radar’s ability to differentiate objects is insufficient
● Radar data is far too noisy to train neural networks
● Radar needs to detect smaller objects from farther away
● Radar is producing false positives, reducing trust in the data
● Radar can’t see stationary objects

These conversations informed our Ago software suite roadmap.

The first step on that path: Spartan captures raw datasets of scenarios described by customers on public roads using our van. We use additional sensors to further validate and provide ground truth for our observations. Then, we draw from a variety of our algorithms to extract a wide range of information from these data sets. This information greatly enhances the understanding of the environment in a complementary way to optical sensors. As soon as we achieve the performance we are looking for, we need to optimize the software to run efficiently on available silicon.

Most modern automotive radar systems forego advanced processing methods due to computational complexity. Spartan instead leverages our extensive systems engineering experience to deploy complex software on limited compute resources.

The slow evolution of radar in the mobility space

In the early 2000s, during the DARPA Grand Challenges, we saw teams of researchers competing in off road autonomous races. Early navigation suites were mainly optical—LiDAR and camera, complemented by GNSS/INS. Most vehicles only made it a few hundred feet from the starting line before crashing or losing track of the path.

Now, nearly 20 years have come and gone, and the promise of autonomy still seems years away. Yes, there may be better integration, but optical sensors are still the leading modality and only a fraction of the information a radar can produce is used for safety or perception. Today many OEMs and Tier 1’s research is often focused on sensors that are too computationally expensive to find their way into full-scale commercialization. For example, interpreting data from a camera takes about 10x the computational power as similar performing radar. Fusing camera and radar data can make a safer and more effective system while preserving resources.

Hundreds of millions of radars already exist on passenger cars today—this technology has proven a key modality for both automatic emergency braking (AEB) and adaptive cruise control (ACC). Most of these sensors don’t deliver the elevation data and resolution that would enable them to play a larger role in other ADAS and AV capabilities. As a result, they provide simple features such as range, working in concert with cameras, where the camera does most of the work. This limited implementation is not fully complementary and does not benefit from all the information that radar can provide, and the performance is still largely limited by camera’s dependency and sensitivity to lighting conditions.

This prevalence of radars represents a huge opportunity to optimize sensor performance and advance safety by adding better software.

Enhancing digital signal processing and bringing focus to radar solutions

Much of our work is focused on the part of the software chain that’s typically hidden from view. Enhancing this software—referred to as the digital signal processing chain—is the key to unlocking radar’s full potential.

Spartan’s software product platform, called Ago, delivers this necessary focus to radar—determining what’s an object and what’s noise. Ago’s main objectives are to enhance radar signal integrity, increase computational efficiency, and give customers the ability to design radar functionality as they see fit. This opens up the opportunity to use the same hardware for multiple detection tasks such as blind spot monitoring on the highway and close proximity sensing in parking lots.

All of these activities are validated using our research van. And, when not in research mode, it’s also a wonderful tool for demonstrations. Few things are more compelling than seeing this software run on public roads in real-time.

Spartan also provides these same sensors to key customers who, likewise, are gathering data using their test vehicles and influencing the development of the industry’s next generation of perception stacks. These arrangements empower customers to gather raw data on specific scenarios, like lane splitting motorcycles or jaywalking pedestrians, and then collaborate with Spartan to develop radar modes that respond to these road events.

This is the essence of software-defined radar development: End users will be able to deploy a huge range of operations on a single, cost-effective sensor—all through software alone. The radar will serve, in essence, as a blank slate to which applications can be added, specific to its operational demands.

This is the future Spartan is delivering.

Cameron Gieda
Vice President of Business Development

Get the Spartan sensing edge.