Status: Active (under active development, breaking changes may occur)
Control a single aircraft (yellow aircraft) to reach the goal position (green star) while avoiding conflicts with other intruder aircraft (red aircraft).
Parameter of the environments can be found in config.py.
The action space is discrete: -1, 0, 1 for change of heading and -1, 0, 1 for the throttle.
The action space is continuous: [-1, 1] for change of heading and [-1, 1] for the throttle.
The input state is the whole image of the map, with four most recent frames stacked together. Action space is discrete.
The environment designed to implement Hindsight Experience Replay algorithm. The observation space is of the following form:
OrderedDict([('achieved_goal', Box),
('desired_goal', Box),
('observation', Box)])
The action space is continuous.
The environment designed to implement Hindsight Experience Replay algorithm with 3/9 discrete actions.
You can use baselines algorithms from OpenAI to solve this problem. Simply run the following:
For discrete action space
cd baselines-master
python -m baselines.run --env=guidance-collision-avoidance-single-v0 --alg=deepq
For continuous action space
cd baselines-master
python -m baselines.run --env=guidance-collision-avoidance-single-continuous-action-v0 --alg=ddpg
Optional arguments:
--save_path the path where you want to save the model
--load_path the path where you want to load the model
--num_timesteps total time steps the agent will be trained
More optional can be found in the documentation of OpenAI baselines repository.
Use Monte Carlo Tree Search Algorithm to solve this problem. Details can be found at the above reference. Run
cd Algorithms/MCTS
python Agent.py