Replace computer vision system with two pigeons.

[TomFraser]

I propose a change to replace the complex and unnecessary machine learning based computer vision system for cone detection with a much more simplistic two pigeon approach. (See attached image)

Advantages of a two pigeon approach:

• Training a pigeon is easier than an ML model
• Pigeons are a readily available resource (can collect local pigeons at track and train before comp)
• Pigeons are much more flexible than a model, they don't care about cone size or lighting conditions
• Bird feed is cheaper than a ZED camera
• Who doesn't like birds???
• The bird stack is

The high level tasks to complete are as follows:

1. Acquire two pigeons
2. Train each pigeon to expect a food reward when it pecks either a blue or yellow cone respectively
3. Build a cage-like structure to house each pigeon slightly angled away from the center line of the car
4. Place a clear touch sensor in front of the pigeons and track their pecks as cone locations
5. Develop an algorithm to determine the best RVWP from the pigeon pecks

Inspired by: https://en.wikipedia.org/wiki/Project_Pigeon

[cwljohnson]

Hey TomFraser, thanks for raising this issue.

Just wanted to flag some security concerns from a potential implementation of this, as this would allow full government access to our vision pipeline and collected data, since all birds (including pigeons) are government drones.

How would you suggest we mitigate this issue?