Doing so marks the first time stereo 3D Simultaneous Location and Mapping (SLAM) was achieved on a drone, where power and weight limitations have until now been stifling.
Being capable of 3D SLAM means that a drone or other vehicle has a detailed understanding of the world around it, meaning implementing autonomous navigation is now easier and accessible to any developer.
Equally significant was the product’s cost: Hardware used in the self-driving cars being developed by Google and others makes use of LIDAR arrays, which cost around $60,000, and budget alternatives are unreliable and provide too little data for such uses. The ZED sensor sees up to 20 meters both indoors and outdoors for only $449, and the Jetson TX1 Developer Kit is just $599. Adding a $3,100 base drone and some basic parts, all the drone’s hardware in the video comes in at less than a month’s rent in San Francisco.
“The single greatest barrier to autonomous navigation is the capability to see and understand,” explains Cecile Schmollgruber, CEO of Stereolabs. “In simulations, computers have been able to navigate around obstacles for a long time. But they’re blind in the real world. With our sensor, a machine can see the world almost as well as a human can.”
Schmollgruber’s comparison with human vision is more than a metaphor. The ZED itself is modeled after human vision. Two cameras (eyes) send video feed to a GPU, where Stereolabs’ software calculates depth maps by measuring the disparity between what it sees, similar to the human visual cortex. The technique allows the ZED to capture depth at resolutions up to 2.2k and frame rate up to 120fps, indoors or outdoors.
How the drone saw the French chateau as it flew around the exterior.
Since its launch this past May, ZED has already attracted buyers from R&D labs in major semiconductor, automotive and robotics companies, in addition to hobbyist developers. The ZED is also being integrated into MIT’s robotics class, and has been used by graduate students since its launch to develop autonomous vehicles that navigate urban environments.
“We believe the ZED has an important future in the autonomous vehicle market,” said Sertac Karaman, Assistant Professor of Aeronautics and Astronautics at MIT. “The ZED camera’s unmatched resolution, frame rate, and its ease of integration makes it a good stereo solution. We have had a great experience with the ZED cameras so far."
To make things easier, the 3D SLAM feature will be added to the existing ZED SDK, giving every developer access to depth, tracking and mapping data. This makes it easy for any machine to understand its position and free space, and eventually it will be easy to choose its future trajectory.
“The ZED camera is a great demonstration of what is possible with the Jetson TX1” said Deepu Talla, VP and GM for Tegra at Nvidia. “The Jetson community of developers is excited by what Stereolabs has achieved here."
The Nvidia Jetson TX1, which provided the graphics processing power for this demonstration, is an embedded GPU platform that provides advanced GPU computing on a module roughly the size of a credit card. Using Maxwell architecture, the platform delivers a full teraflop of performance with minimal power requirements, with special features designed to support artificial intelligence and machine learning.
The ZED depth sensor is available for $449 at www.stereolabs.com. The Jetson TX1 Developer Kit was announced last Tuesday and is available now for $599.99. The production-ready Jetson TX1 module will be available early next year at $299.