fire-drone

Description

fire-drone is a project to develop a certifiable AI system for an autonomous fire-fighting drone.

Project Organization

The project is organized into 4 main submodules. Each submodule contains directories for continuous and discrete components. Some of the submodules may be missing one of these components if no files were created for that components.

The linkages between the various submodules is illustrated below:

All of the modules on the lower level are combined together within the experiment submodule.

• agents/ - collection of agents that interact with the environment.
  • discrete/hopper.py - a discrete agent that 'hops' between locations in the enivornment.
  • continuous/drone.py - a continuous agents whose dynamics are defined by [#wang2017safe].
• environments/ - collection of environments and related models
  • discrete/hyper_plane.py - a model for an n-dimensional grid-like environment.
  • discrete/fire_propagation.py - a model for convolution based fire propagation.
• experiments/ - definitions for running experiments.
  • discrete/experiments.py - defines the three main algorithms for running. experiments: pre-, peri-, and post- within a class structure for ease of testing againts various agent/environment configurations.
• lib/ - directory of additional helper functions.
  • helper.py - a collection of useful auxilary functions.
  • plotting.py - functions for plotting data output by the policy.
• policies/ - container for various policy implementations.
  • discrete/actor_critic.py - an implementation of Actor-Critic architecture that can output 1D, 2D, or 3D coordinates.
• dryvry_verify.py - contains the TC_Simulate that will be called by DryVR to simulate the dynamics of the drone
• main.py - main script to run the project

Installation/Setup

To be able to run the project, first install DryVR following the instructions located here in a location of your choosing. Next, clone this directory inside the DryVR directory. Finally, use pip to install all necessary requirements:

pip install -r requirements.txt

Usage

To run the the project, navigate into the src/ directory and execute

python main.py

This will run the experiments defined inside main.py. To run other experiments, simply modify the contents of this file.

Additional Notes

This section contains a list of importants notes about the project.

• The dryvr_verify.py file has to be in the same directory path as main.py in order for the static variable Drone.GOAL_POS to contain the correct value. If the file is not in the same level, DryVR will consider the Drone class inside the TC_Simulate function to be different from the Drone class used by main so the value of Drone.GOAL_POS will be incorrect.

• The agent hopper.py was subsumed by drone.py and may not be complete. Initially, hopper.py was used to drive the exploration process around the environment for faster processing but it became necessary to later verify the path of the agent as it actually moved around in the environment so this class ceased to be used. In some modules, you may see train_agent and verify_agent variables because of this reason.

#wang2017safe