In this project, DRL was used to train an autonomous quadrotor to be able to fly in a closed space, eg, a room. However, the trained policy did not perform well but the reasoning is already mentioned in the dissertation report. Moreover, as a follow up, I used preimplemented algorithms, in case my implementation was wrong, and also changed the reward mechanism. This changed the performance of the quadrotor, as it was able to sustain a flight within the constrained space, without colliding. The code is yet to be uploaded, but will update here.
This repo contains code and dissertation report from my Master's course. Code relevant to training the policy and testing the performance has been added, but some irrelevant part has been omitted. However, should someone feel that the code in incomplete, please raise an Issue and I'll see what I can do.
The dissertation report should also be available in the same repo.
Good luck.