Controller Optimization Using Deep Learning

Deep learning for controller optimization.

Click for Youtube video:

Prereq.s

Tensorflow

python libraries:

chmod +x pythonReady.sh
yes "yes" | sudo sh pythonReady.sh

run

python main.py

ToDo

• Implement NN in Propeller
• Try RNN
• Try Q-Learning to optimize the controller performance

Problem Setup

Stage 1: Show NN can perform PID less than equal to the performance of PID controller.
Stage 2: Show Q-learning can optimize NN structure.

Results

MSE = 1.6

Red: Actual PID output for ESC signal
Blue: NN output for ESC signal