At Nvidia‘s GTC conference last week, Somerville, MA-based stereo-vision technology startup Nodar and high-definition camera and AI-based imaging solution company Leopard Imaging demonstrated advanced 3D vision capabilities coming from their new strategic partnership. The collaborators aim to provide customers with enhanced perception capabilities, demonstrating accurate depth mapping and point-cloud generation at ranges significantly exceeding current market solutions at up to 200 m (656 ft).

“By integrating Nodar’s Hammerhead software with Leopard Imaging’s high-performance cameras, we are redefining 3D vision for autonomous and robotic systems,” said Dr. Leaf Jiang, Founder and CEO of Nodar. “Our stereo vision delivers high-fidelity depth maps up to 1000 meters while maintaining exceptional accuracy in heavy rain, dense fog, and low light. This partnership demonstrates how advanced stereo vision can surpass lidar, offering superior perception at a fraction of the cost.”

The collaboration integrates Nodar’s Hammerhead software with Leopard Imaging’s GMSL2 cameras, including IMX031-based solutions with Nvidia Drive AGX Orin compute. It focuses on serving key markets, including last-mile delivery, trucks, and robotaxis, with Leopard Imaging leveraging its extensive distribution network in both the U.S. and Europe.

“At Leopard Imaging, we are committed to providing state-of-the-art imaging solutions that enhance vision-based systems across industries,” said Bill Pu, President and Co-Founder of Freemont, CA-based Leopard Imaging. “Our collaboration with Nodar brings together best-in-class stereo vision software and advanced camera hardware, creating a powerful solution for customers seeking superior 3D perception.”

 

Going deeper on technology

At GTC, Futurride was briefed on the collaboration by Cliff Cheng, Senior Vice President of Marketing at Leopard Imaging, Inc., and Barrs Lang, Chief Commercial Officer at Nodar.

For a small startup like Nodar, “Leopard represents business channels that we don’t have today,” said Lang. “Leopard’s been around, they have a big reputation in the market, and they make hardware, deliver it, and support the customer in the field. If we can enable them to license in the software, then I think it’s a great win-win.”

As an IATF 16949-certified vendor, Leopard currently supplies commercial vehicle applications like trucks, delivery vans, and delivery robots. With the Nodar partnership, Leopard can extend its customers’ perception from short to long range, said Cheng.

“Today, a lot of our customers are using lidar for long range,” said Cheng. “This is a good substitute product for the lidar. You still need the camera system to tell ‘that’s a person, there’s an object.’ You need to classify. With this solution, two cameras can do all that already.”

Most stereo cameras in prototype and production vehicles have 5 to 10 cm lateral spacing, according to Lang. As that spacing is increased, triangulation measurement can significantly improve accuracy for long-range imaging.

“Our technology enables the widening of the baseline of a stereo system, which then allows for significantly longer range,” he said. “Most stereo cameras max out at 6 to 8 m; for us, we’re talking about 600 to 800 m. That’s customizable based on the use case.”

The company’s software products enable this performance leap. In addition to Hammerhead for calibration and stereo matching modules, the output being a point cloud, GridDetect perception software uses Hammerhead point cloud and produces the bounding boxes and velocity information that is the next layer of the perception stack.

 

Better perception at much lower cost

Nodar’s ultimate aim is to help its customers reduce costs and improve durability for high-performance autonomy systems by replacing lidar with stereo cameras and software.

“What we’re seeing is that lidar is a necessity right now for most customers that are deploying high-end perception stacks,” said Lang. “But they’re not happy with the solutions, they’re expensive. Spinning lidars have a one-year MTBF. Effectively, they’re not meant for 24/7 operation.”

Cameras are designed to be installed for a decade, so the two technologies have different reliability and long-term deployment capabilities, he emphasized.

“For us, customers that are using lidar are seeing that our output can supplement or replace that output for the same use at a significant cost savings because we’re talking about the software on top of the camera costs, not a whole different sensor that costs thousands of dollars,” said Lang.

Those customers are coming not just from the passenger car industry.

“We find that, [for] off-road vehicles, there’s been less investment from the lidar companies,” he said. “We compete with Iidar in that we produce long range point cloud data, and so for us having 50 million points a second of resolution allows us to be in a lot of different environments outside of just on-highway.”

One of those could be rail, for which the Nodar technology could enable seeing down train tracks out to a kilometer.

“You can see a 50 by 50 cm target at 600 plus meters away,” he concluded. “That truly is just about the technology of how good cameras are for us.”

 

Latest development kit

Earlier this month, Nodar announced its new Hammerhead Development Kit (HDK) 2.0 to provide better access to its ultra-wide baseline, untethered stereo 3D vision system. The latest development kit enables rapid evaluation and easy integration of Nodar’s technology, streamlining the deployment of long-range object detection, high-resolution depth perception, and exceptional accuracy in real-world conditions.

The system allows engineers to integrate Hammerhead’s real-time, high-fidelity 3D point clouds into their applications along with industry-best object detection via GridDetect. It includes all necessary hardware—cameras, lenses, mounts, compute modules, and cables—and can be set up in minutes.

Nodar technology is designed to be deployed in a range of applications, including AVs (autonomous vehicles) and ADASs (advanced driving assistance systems) in the automotive space, as well as rail, heavy equipment, agriculture, construction, security, surveillance, and aviation. For instance, TDK is leveraging the HDK to enhance its positioning and SLAM (simultaneous localization and mapping) solutions.

“Nodar’s technology offers us better resolution, accuracy, and range, significantly enhancing our positioning and localization applications,” said Chris Goodall, Managing Director and President at Trusted Positioning Inc., a TDK Group Company. “Unlike LiDAR, which lacks color data and has limited range and resolution, Nodar’s system provides more detail and enables us to offer our customers better performance at lower cost. For urban driving localization, Nodar’s higher vertical field-of-view enables us to identify unique features that LiDAR cannot detect. When added to localization for systems with GPS plus radar, Nodar elevates the overall accuracy of the system.”

Specifically, the Nodar technology in the HDK 2.0 is said to offer several distinct advantages over traditional lidar and other sensing technologies. It can detect a 1 m tall object at 1000 m with 1.8% accuracy.

Sensing at 5 megapixels per frame and 50 million depth measurements per second provide reliable, high-fidelity 3D mapping with true depth per pixel, eliminating interpolation errors. It is said to operate effectively in low light, fog, rain, dust, high vibration, and direct sunlight. Auto calibration actively compensates for angular changes of cameras of up to 0.05 degrees per frame caused by vibration, engine noise, temperature shifts, and other system shocks, maintaining the integrity of depth measurements. The use of off-the-shelf cameras makes Nodar technology more affordable and robust than lidar.