What I Do: Making Radar See What It Naturally Struggles With
Radar is excellent at certain things. It detects object velocity with precision, sees in complete darkness, and works through rain, fog, and snow. However, its natural angular resolution is low, so it struggles with spatial separation between targets. In my role as Algorithm Team Lead at Arbe, my team and I develop algorithms that overcome these limitations. We improve the radar image across multiple dimensions: better target separation and detection, better detection of weak and almost invisible targets, clearer object representation in the point cloud, and more. The exciting part is creating algorithms sophisticated enough to genuinely improve spatial resolution and dynamic range, while keeping them simple enough to run in real-time on limited radar processors.
The Process: Researching New Algorithms for Image Improvement
The algo team is composed of experienced people, who love a new challenge. We identify the core issues that limit the radar image quality, and search for the cutting edge algorithms to solve them. The hard part is converting an algorithm to fit the specific form of the radar data without damaging the algo’s capabilities. This requires long-term research and multiple attempts, using both simulated and real life data.
The Challenge: Balancing Sophistication With Real-Time Constraints
The best algorithm is useless if it can’t run fast enough in an actual vehicle. Once we have identified a solution, we have to convert it to a form that will work under the tight processor constraints. We need solutions that deliver real improvements to Arbe’s high definition radar image quality while running on hardware with strict power and processing limits. That means making tough choices about which improvements matter most, optimizing code paths, and sometimes finding creative shortcuts that give us most of the benefit at half the computational cost. We work closely with the programming team, who write directly to the radar software, in order to achieve these goals. The result is radar perception that compensates for radar technology’s natural weaknesses, delivering spatial clarity that matches its already strong velocity detection and all-weather performance.
