On the PHANTOMBots package we provide several tools for Multi-Agent Pursuit Evasion.

On the environment directory several tools are available to run simulations under different constraints.

• env.py contains the class Environment which holds all the functions needed for general simulations. Some noteworthy functions are:
  • win_condition returns true if the target robot is on the same position as a pursuer.
  • play_round takes in the actions for all bots and moves them
  • adjacent gets all adjacent positions to an XY position
  • legal_move_pos checks if given movement is legal (on the grid and no collisions)
  • animatemakes a video of the current simulation history positions
  • several other functions can be found
• game.py contains the class Game which runs simulations and returns training samples for ML
  • get_simulation_history runs a simulation and returns actions, states and rewards.
• phantomBot.py contains the class PhantomBot which represents a single bot
  • move changes the XY position of the bot
  • double_move performs two movements to abstract speed

On the CV directory the relevant files to run the physical system are present. For setting up robots refer to Paperbot. There are several files which we will not go into detail.

• config.py contains configuration info like ip addresses, movement speeds, sleep times, etc.
• RunProject.py contains core functions for running the system.
  • run_system is the function that initializes everything. The instructions to simulate are passed here.

For an example on running a system refer to physicalRun.py on the main directory.

On the ML directory we find the machine learning functions and classes.

• tf_reinforcement.py contains the ML model used, implemented in TensorFlow.
• tf_simulator.py is used to run and train using the file above.

For an example of usage, refer to learn_simulate.py.

On the baseline directory 3 different baselines are available.

• runner.py contains BaselineRunner, a class used to run simulations on any of the baselines available.