Control Methods

Collection of modern control strategies implemented in:

• C++, using Eigen and osqp-eigen
• Python, using numpy and cvxpy

State Space Model

For a given LTI state space model, two controllers were implemented:

• Model Predictive Control: Nominal MPC controller with input constraints.
• Data-enabled Predictive Control: DeePC controller that constructs its input and output Hankel matrices once instantiated.

Markov Decision Process

For a Markov Decision Process, such as Tic Tac Toe or a grid with a starting cell and goal cell, a few controllers were tested:

Model-dependent

Policy Iteration and Value Iteration were used to generate a policy for the processes given their models' transition probabilities.

Reinforcement Learning

Some variations of reinforcement learning were used to obtain efficient model-independent policies for the processes, namely:

• Q-Learning
• State–action–reward–state–action (SARSA)
• Monte Carlo Learning