Zadar Labs, a leading developer of 4D imaging radar, is building a first-of-its kind deep learning training dataset for radar-based perception. Zadar is accelerating the path toward radar-based target classification through a collaboration with Xilinx, Inc. to develop and leverage its 4D radar deep learning processor unit (DPU).

Radar is a robust and widely utilized technology in many industries, most notably automotive ADAS and autonomous vehicles and trucks. The ability to perform target classification using radar is expected to significantly expand the capabilities and applications of safety and automated systems.

Zadar’s unique approach of developing an end-to-end Software-Defined Imaging Radar (SDIR) platform generates the high quality point clouds needed to build the training dataset for deep learning algorithms. Zadar and Xilinx have identified a gap in the market when it comes to imaging radar. Having better data is only part of the solution. Data intelligence and insights are needed to truly advance radar performance. Other mainstream sensor modalities, such as camera and Lidar, have an ecosystem of artificial intelligence (AI) and machine learning (ML) approaches that best suit their respective data types. However, these established techniques do not readily lend themselves to imaging radar.

The depth and quality of the imaging point cloud provides not only detection location, but velocity as well as signal-to-noise ratio and the power of the reflection. Zadar’s SDIR solution takes this one step further by including additional detection characteristics. Fundamentally, new AI and ML techniques are required to bring out the untapped potential of imaging radar.

Currently, the team at Zadar is focusing on a training dataset for target classification. However, this is only the beginning. As the dataset is refined and scales so too will the opportunity for perception engineers to develop novel radar-based algorithms.

Zadar’s SDIR creates a layer of intelligent operation on top of 4D imaging technology as well as provides access to data along the signal and data processing chain. This enables the platform to be dynamic and work in collaboration with the perception algorithms to respond to the operating environment and provide optimal performance based on the needs of the system. For instance, if the scenario requires, Zadar’s solution can discern the movement of a human’s arms relative to the body or identify the wheels of a vehicle. It is this level of performance that is required to achieve radar-based target classification.

Zadar’s dataset is a powerful tool in building the foundation of radar-based perception. Through their collaboration, Xilinx has utilized the dataset for the development of their 4D radar DPU.

“We developed our 4D imaging radar platforms to bring the power of radar vision to the world and unlock deep learning and AI applications for radar and radar fusion with other sensors.” said Mahmoud Saadat, CEO and co-founder of Zadar Labs. “We are excited to continue our collaboration with Xilinx to bring the most advanced radar vision solution to market.”

“Our collaboration with Zadar Labs represents an important milestone for 4D radar-based perception,” said Willard Tu, senior director, Automotive, Xilinx. “Together we will work to provide automotive OEMs and Tier-1 Suppliers with a solution that will expand the capabilities of safety and automated systems. We look forward to working closely with Zadar Labs to equip the auto industry with this innovative approach that will provide enhanced vehicle data intelligence and insights.”