Next project: use yozh together with a cheap AI camera from DFrobot, Huskylens.

Task: find a "container" placed on the field, read the tag on it (AprilTag), pick it up and deliver to one of locations depending on the tag.

I already had working circuit python library for Huskylens, so I used that.

But there was a number of gotchas: 

• Turns out tof distance sensors readings are much higher than real life distance if the object is not at the center of the field of vision but on the periphery. 
• Huskylens can use one of two interfaces: I2C or UART. By default, it is set to "auto", selecting interface depending on the incoming signal. It doesn't really work; if you want reliable operation, you need to select the correct interface in general settings.
• Huskylens can recognize apriltags reasonably well - if the camera is stationary. Trying to recognize the tag while, e. g. the robot is turning doesn't work.

Thus, I used mixed strategy: using distance sensors for initial detection and approaching the object; once we are close, use Huskylens to read the tag and do the final course correction before picking it up.

Here is the final result: