Occipital Announces Availability of Structure Core 3D Sensor

Depth sensing and 6-DoF spatial awareness for $400

4 min read
Misty robot
Occipital’s Structure Core sensor on the Misty II robot.
Photo: Occipital

When Microsoft’s Kinect came out, it’s probably fair to say that it revolutionized robotics. As soon as folks figured out that they could get halfway decent 3D vision for cheap, Kinects started to get kludged on to every robot that moved (or didn’t), even if that robot already had a much fancier and more expensive 3D vision system on it (I’m looking at you, PR2). But Kinect was a gaming sensor—not only was it not optimized for robotics even a little bit, Microsoft seemed to be not all that interested in supporting the robotics industry, and as the Kinect got older, people were forced to lurch painfully over to whatever else happened to be available, like PrimeSense (acquired by Apple) or Asus Xtion sensors. Or, they just stuck with the Kinect, which you still see robots using today.

There’s clearly a need for high quality, high performance, affordable 3D sensors designed for robotics applications, and today, Occipital is announcing pricing and availability for their Structure Core sensor—a high precision, self-contained 3D perception system that goes on sale early next year for US $400.

As you can see from the video above, there are robots (Misty II is one of them) already using the Structure Core sensor.

The device is based around a pair of infrared cameras and a high contrast IR pattern projector that Occipital assures us won’t laserify anyone’s eyeballs but will still work outdoors in direct sunlight. The cameras are synched to a global shutter (as opposed to a rolling shutter), ensuring that you’ll get good data even of objects that are moving. An additional RGB camera gives you registered RGBD images, or you can opt for an ultra-wide angle visible light camera instead. There’s also an IMU on board to enable 6-DoF positional tracking, and the whole thing is backed up by a new SDK for generating things like point clouds and depth maps. You’ll need a USB 3 connection to properly take advantage of all this, because there’s too much data for USB 2.

If you’ve read this far, here are the detailed specs you’ve probably been waiting for:

Occipital Structure Core Specs

We’re told that Structure Core sensors will be shipping in March 2019 for $399, but if you want one sooner, you can pay a little extra for early access pricing—$699 gets you a sensor within the next two weeks, and $499 will get you one in January. And if you want a whole pile of sensors for that robot army we know you’re building, ask Occipital for volume pricing. 

Occipital Structure Core sensor on a drone Structure Core sensors on a drone. Photo: Occipital

For a bit more detail, we spoke with Occipital CEO and co-founder Jeff Powers via email.

IEEE Spectrum: Why should roboticists in particular be excited about Structure Core?

Jeff Powers: We think roboticists face enough challenges building robots—from compute to actuation to charging—that anything we can do to make things run better is a net good. Being able to just plug in rich 3D perception over USB, all synchronized and paired with a great SDK, is a massive reduction in time and cost to bring a functional spatially-aware robot into existence.

“Imagine you’re a kid 20 years from now building a robot—you’ll almost certainly just have some kind of ‘vision cortex’ you can plug into your robot, making it environmentally aware. Structure Core is the precursor to that”

How is Structure Core different from other sensors on the market?

Structure Core combines ultra-wide visible spectrum camera for 6-DoF tracking, dense depth for scale and 3D scanning, and a fully-synchronized IMU to tie it all together. This is the stuff you find on things like the Hololens, Magic Leap, and indoor robots. Generally speaking, other 3D sensors on the market only include a portion of this, meaning you still have to integrate multiple systems together to make a fully functional spatially-aware product.

Specifically, can you compare Structure Core to Intel’s RealSense?

Intel’s RealSense D415/435 are close, but due to their specs, it’s a little more tuned for 3D scanning and less so for the 6-DoF spatial awareness we’ve aimed Structure Core at:

  • Structure Core’s depth precision is higher than both Intel sensors (dramatically higher than the D435), in part thanks to a wider distance between cameras. Core’s depth FOV is similar to the D415, but narrower than the D435.
  • Structure Core has all-global shutter cameras (both depth and visible), which is important for real-time computer vision, especially when you’re dealing with fast motion.
  • Structure Core has a built-in IMU standard across all variants (Intel just started offering an IMU variant for the D435 only.)
  • Structure Core’s visible camera has two options: A 160-degree ultra-wide visible camera for tracking, or an 85-degree RGB camera, both with global shutter. RealSense has just one 77-degree RGB camera option, but with a 1080p resolution rather than the 480p in Structure Core.
  • Structure Core is thinner and lighter (about 28 percent lighter and 25 percent thinner than D435), though is slightly longer.

What kinds of applications will this sensor enable that weren’t possible before?

Structure Core allows smaller teams, independent developers, and researchers to bolt in 3D perception. The tech giants and mega-funded startups can afford to build or acquire 3D sensing and have huge computer visions teams, but Structure Core will help democratize the ability to build a spatially-aware product. Imagine you’re a kid 20 years from now building a robot—you’ll almost certainly just have some kind of “vision cortex” you can plug into your robot, making it environmentally aware. Structure Core is the precursor to that.

[ Occipital Structure Core ]

The Conversation (0)

How the U.S. Army Is Turning Robots Into Team Players

Engineers battle the limits of deep learning for battlefield bots

11 min read
Robot with threads near a fallen branch

RoMan, the Army Research Laboratory's robotic manipulator, considers the best way to grasp and move a tree branch at the Adelphi Laboratory Center, in Maryland.

Evan Ackerman

“I should probably not be standing this close," I think to myself, as the robot slowly approaches a large tree branch on the floor in front of me. It's not the size of the branch that makes me nervous—it's that the robot is operating autonomously, and that while I know what it's supposed to do, I'm not entirely sure what it will do. If everything works the way the roboticists at the U.S. Army Research Laboratory (ARL) in Adelphi, Md., expect, the robot will identify the branch, grasp it, and drag it out of the way. These folks know what they're doing, but I've spent enough time around robots that I take a small step backwards anyway.

This article is part of our special report on AI, “The Great AI Reckoning.”

The robot, named RoMan, for Robotic Manipulator, is about the size of a large lawn mower, with a tracked base that helps it handle most kinds of terrain. At the front, it has a squat torso equipped with cameras and depth sensors, as well as a pair of arms that were harvested from a prototype disaster-response robot originally developed at NASA's Jet Propulsion Laboratory for a DARPA robotics competition. RoMan's job today is roadway clearing, a multistep task that ARL wants the robot to complete as autonomously as possible. Instead of instructing the robot to grasp specific objects in specific ways and move them to specific places, the operators tell RoMan to "go clear a path." It's then up to the robot to make all the decisions necessary to achieve that objective.

Keep Reading ↓ Show less