ICYMI AI for visual object detection (and tracking) has become a robotics “must have”.
My approach has been the addition of a stereo camera with a neural processor, from the Oak-D family.
Recently the Raspberry Pi Foundation announced the Hailo AI Hat for Raspberry Pi which provides up to 13 TOPS neural processing for $85
What does it enable?
That’s a mnumonic that I’m not familiar with.
(TAMTINFW) 
It enables a video that takes forever to load.
(I bailed on it after 2 minutes on a 4G-LTE connection that loads 15 minute YouTube videos in seconds.)
First comment in the list of comments:
“Another company that’s not up-front about their prices” - apparently they’re one of those “contact us” for prices firms.
Me - if they’re not forthcoming with their prices, I assume it’s too expensive and move on.
“In Case You Missed It” - 2016 Article:
Yeah, weird stuff the link.
The Official Raspberry Pi AI Hat link:
US purchasers:
I wouldn’t trust:
and I think a 1.3 version PiCam as well.
The only real problem is that you’ll need a small lead weight on the bottom of the robot to keep it from being top-heavy. 
Are you thinking of getting one? It might be useful coupled with the Pi-5.
Price isn’t that bad. If I was playing with that stuff, I’d be tempted!
No. While the RPi AI kit is 13TOPS, I have not seen it applied to dual PiCams for 3D pointclouds yet, or ROS and RTABmap visual Localization and Mapping.
Whereas the Oak-D cameras already have the ROS drivers and RTABmap configuration files deployed.
