Dodge-the-Blocks

Dodge-the-Blocks is an experimental project, i created to learn reinforcement learning algorithms on customizable environment, as beginner(holds true for today also) i was curious how the behaviour of agent will change if number of states available to it change and how does agent learning pattern agent in a markov decision algorithm in Q-Learning algorithm

The game support users also to play themselves but it becomes very hard unless you make some bot or something to score past 15 i think! The current setup for agent has only two types of state defined for it

1. three-state-system : In this state detection approach, state of agent is decided depending on relative posistion of safe position wrt to agent's current position and accordingly a state is assigned to the agent

1. `0` (Safe block is left to current position of agent)
2. `1` (Safe block is just above the current position of agent)
3. `2` (Safe block is right to current position of agent)

Depending upon what position the agent is wrt to agent should figure out what should be the best action for it to take(move left, right or stay) *Update (April-2024):*: Updated reward system shows that agent is able to learn 3 state system game in 1 episode this required making gamma 1 so future states value is equal to current state and some minor modifications in reward system. 2. twenty-five-state-system : This approach is more detailed and has about 25 state to create search space for learning for agent, the states in this approach are defined wrt abosulte 5 positions that agent can take and 5 possible safe blocks(resulting in 5\*5 = 25 states). game/environment is always aware of the position of safe block but agent is not, agent is only aware of the state and has learn/adapt to make decisions to increase the score for each episodes

Setup

If you want to give it a try you can clone the project

      git clone https://github.com/indiVar0508/Dodge-the-Blocks.git

Install the requirements into your Python virtual environment, make sure you are in Dodge-the-Blocks directory

    python3 -m venv venv # create virtual environment
    source venv/bin/activate # activate virtual environment in terminal
    pip install -r requirements.txt

how to run?

you can check CLI option to run from Terminal to run game in various mode (play yourself vs let the RL agent play) use the command python dodge_the_blocks/main.py -h to check options available

    Run Game as per User's preference

    optional arguments:
      -h, --help            show this help message and exit
      --use_agent use_agent  specify state for agent to play in 3 state vs 25 state

To run game for yourself use run the command python dodge_the_blocks/main.py, if you want to let agent learn and play the game use optional argument --use_agent which only expects 3 or 25 as valid states for agent