Tangram Vision Promises "Plug-'n'-Play" Depth Sensing and Edge AI Smarts with its HiFi 3D Sensor

With 1.6 megapixels of depth data from 2.2 megapixel global-shutter cameras and 8 TOPS of AI compute, the HiFi is a miniature marvel.

Tangram Vision is looking to make it easier to add depth-sensing and computer vision to projects, launching a crowdfunding campaign for the HiFi 3D Sensor — promising "plug-'n'-play" capabilities for robotics acceleration for edge artificial intelligence (edge AI) on-device.

"We pulled out all the stops to make HiFi the most powerful and performant 3D sensor possible for under $3,000," claims Tangram Vision's Adam Rodnitzky of the company's flagship product. "It features a cutting-edge TI [Texas Instruments] processor, high-quality optics, and multiple convenience features (like both USB-C and Ethernet ports) to raise the bar for what to expect from a high-quality depth sensor."

Tangram Vision is hoping to outdo rival 3D sensors costing an order of magnitude more with its compact yet powerful HiFi. (📹: Tangram Vision)

The sensor is built around a custom optics system which, the company claims, delivers 1.6 megapixel of supersampled depth data from 2.2 megapixel global-shutter camera sensors and a pair of laser pattern projects while also delivering a 136° field of vision. Its accuracy, according to in-house testing, hits a sub-one per cent depth error from one foot to over 16 feet. The sensor is also claimed to be self-recalibrating, automatically healing misalignments during operation.

While all this is available by connecting the device to an external system via USB or Ethernet, Tangram Vision also promises support for on-device workloads — including machine learning and artificial intelligence. "Inside HiFi, you'll find a powerful, feature-packed Texas Instruments Jacinto processor [with] a dedicated deep-learning matrix multiply accelerator (MMA) that runs at up to 8 TOPS (tera-operations per second) at 1.0GHz," Rodnitzky explains. "And we've paired that MMA with 8GB of onboard memory, so you can load and run sophisticated neural nets right on HiFi."

To prove its point, the company has released demos which use TI's Deep Learning Library, TIDL and other libraries, to provide multi-target pose estimation, 3D scene segmentation, object identification, 3D feature tracking, and AI-enhanced depth maps, all running in real time on-device. Tangram Vision has also showcased the sensor's support for the Robot Operating System 2 (ROS 2), offering plug-and-play streaming of ROS 2 topics out-of-the-box.

The HiFi 3D Sensor comes with the promise of high-quality depth mapping — with the lag in this demo blamed on the streaming of uncompressed video, rather than pure depth data. (📹: Tangram Vision)

While the HiFi may be Tangram Vision's first crowdfunding campaign, the team has experience: "We were part of the team that launched the Structure Sensor 3D sensor right here on Kickstarter ten years ago," Rodnitzky writes, referring to a 3D-scanning tablet accessory which raised over $1.2 million in 2013. "We learned a lot of lessons from having shipped tens of thousands of Structure Sensors, and we’ve used those as part of the foundation from which we're building HiFi."

The HiFi 3D Sensor is now funding on Kickstarter, with Super Early Bird backers able to order a unit for $349 — a claimed 36 percent discount on its eventual retail price. Hardware is expected to ship in April next year, Tangram Vision has confirmed.

Gareth Halfacree
Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire: freelance@halfacree.co.uk.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles