Would I be able to use just the AI deck to make a drone fly through a door by itself?

[cuber-1]

Would this be a possibility?

[gemenerik]

This must be possible, but it is not trivial at all and would involve designing your own or porting an existing neural network / CV algorithm to detect the door. That said; to give you some pointers you could take a look at monocular depth estimation, optical flow estimation, gate/gap detection.

[cuber-1]

Thank you. I was also wondering how I would go about commanding the drone to say, "go backwards" if the camera detects that it is close to a wall or something. I am on Windows OS, by the way.

[cuber-1]

Additionally, would a set frame rate be necessary for a project like this?

[cuber-1]

Do you recommend any specific datasets?

[ToveRumar]

Obstacle avoidance is also not a trivial functionality of the AI-deck. If you do not need the algorithm to be very sophisticated, this can be done by setting a new set point via CPX. Please see the following documentation for pointers;
https://www.bitcraze.io/documentation/repository/aideck-gap8-examples/master/other-examples/stm_gap8_cpx/
https://www.bitcraze.io/documentation/repository/crazyflie-firmware/master/functional-areas/crtp/crtp_commander/
https://www.bitcraze.io/documentation/repository/crazyflie-firmware/master/functional-areas/sensor-to-control/commanders_setpoints/

I would say a fixed frame rate is always easier to work with, however a non fixed could probably also work fine for the avoidance part.

Reagarding the datasets for this, I would suggest looking into the re-search that has been done using the AI-deck, in this blog category https://www.bitcraze.io/category/ai-deck/

I hope this could point you in the right direction.