Skip to content

Latest commit

 

History

History
52 lines (32 loc) · 3.01 KB

README.md

File metadata and controls

52 lines (32 loc) · 3.01 KB
Raw

FerryGym

A gym environment to simulate a ferry traversing the Kiel fjord.

Description

This environment should simulate a ferry agent navigating in the Kiel fjord. The goal is to reach a target location while staying on the fjord and avoiding collisions with other ships.

FerryGym

Action space

The ferry can accelerate (or brake) and turn its heading. Acceleration is measured in m/s^2, and limited to physical possible values. Turning is measured in degrees.

Observation space

The observation space is a dictionary containing the following keys: - agent_position: A 2D vector representing the position of the agent in environment coordinates. - agent_direction: A floating point value representing the direction the agent is facing in degrees. - agent_speed: A floating point value representing the speed of the agent. - target: A 2D vector representing the position of the target in latitude and longitude coordinates. - neighborhood: A 3D array of unsigned integers with a shape of (NEIGHBORHOOD_SIZE, NEIGHBORHOOD_SIZE, 3) representing the agent's local environment. - agent_starting_position: A 2D vector representing the starting position of the agent in latitude and longitude coordinates. NOTE: This environment can be used to generate the training set for the PPUU algorithm. Use the 'generate_training_data' flag to enable this mode. This will save a new pandas dataset with an additional column containing the path to a file containing the neighbourhood for each ship and timestep.

Episode termination

The episode ends if one of the following occurs: 1. Termination: the agent reaches the target. 2. The agent is out of the fjord. 3. The agent collides with another ship. 4. If a dataset of ship trajectories is provided, the dataset is exhausted.

Rewards

The agent receives the following rewards: - running: if nothing happens, the agent receives a default reward of 0. - dataset_exhausted: if the provided dataset is exhausted, the agent receives a default reward of 0. - success: if the agent reaches the target, it receives a default reward of 100. - collision: if the agent collides with another ship, it receives a default reward of -100. - out_of_bounds: if the agent leaves the fjord, it receives a default reward of -100.

Start and target position

The starting state is the ferry dock Reventlue (N54°19.961' E10°09.190 | 54.332683, 10.153167), the target is the ferry dock Moenkeberg (N54°21.117' E10°10.587' | 54.351950, 10.176450)

Installation

Prerequisites

git clone
cd FerryGym
poetry install
poetry run python ./ferrygym/play-env.py

Notes

  • Poetery fails to instal jupyter since it is failing to install debugpy. (At least on my machine) Workaround: poetry config installer.modern-installation false see issue and workaround