From 232299eccff1e9575973d17c6e2a615781b84ebe Mon Sep 17 00:00:00 2001 From: Saquib Nadeem Hashmi Date: Sat, 3 Jul 2021 14:26:42 +0530 Subject: [PATCH] Added README.md --- README.md | 6 ++++++ 1 file changed, 6 insertions(+) create mode 100644 README.md diff --git a/README.md b/README.md new file mode 100644 index 0000000..af2f0ab --- /dev/null +++ b/README.md @@ -0,0 +1,6 @@ +# Automated Quadcopter Flight + +### **This repository contains the code for the research done on automating the flight of a Quadcopter using Deep reinforcement learning. The link to the reseach paper is:** [Performance comparison of deep and shallow network for quadcopter automation](https://ieeexplore.ieee.org/abstract/document/8721383/) + +Traditional learning approaches proposed for controlling quadrotors or helicopters have focused on improving performance for specific trajectories by iteratively improving upon a nominal controller. In these schemes, however, it is not clear how the information gathered from the training trajectories can be used to synthesize controllers for more general trajectories. Motivated by the generalization capability of deep learning, here we try to synthesize control for trajectories using neural network-based model. To test this, we will assign the quadrotors to a random position and observe how it perform to reach the target with emphasis on reward values and change in its positions and velocities. Here we demonstrate the performance of our on 4 tasks: Take-off, Hover, Landing and all 3 combined +