Science Score: 44.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
✓CITATION.cff file
Found CITATION.cff file -
✓codemeta.json file
Found codemeta.json file -
✓.zenodo.json file
Found .zenodo.json file -
○DOI references
-
○Academic publication links
-
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (5.0%) to scientific vocabulary
Repository
A 2d scenario for rl demonstration
Basic Info
- Host: GitHub
- Owner: Fsn9
- Language: Python
- Default Branch: main
- Size: 141 KB
Statistics
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
- Releases: 0
Metadata Files
README.md
rl2d
This code launches a Python Tkinter interface displaying a learning process of a Q-learning agent in a 2D grid-world.
Two navigational tasks are available for learning, each with its own environment: * Learning to reach a point in an empty world; * Learning to reach a point in an obstacle scattered world.
Run the simulation
To run with default parameters, enter in the terminal: python run_rl2d.py
Parameters
Q-learning:
--learning_rate, default=$0.1$. The learning rate $\alpha \in [0,1]$.--discount_factor, default=$0.99$. The discount factor $\gamma \in [0,1]$. In the extreme, for $\gamma=1$ we have a long-term view agent. For $\gamma=0$ we have a myopic agent.--episodes, default=4000. The number of learning episodes $e$.--initial_epsilon, default=1. The initial epsilon $\epsilon_i$ is the exploration probability in the beggining of the learning process. An $\epsilon = 1$ represents a totally random agent. An $\epsilon = 0$ represents a totally greedy agent.--final_epsilon, default=0.05. The value of the final exploration probability $\epsilon_f$.
Environment:
--env_type, default='empty'. The type of the environment $e_t \in {empty, obstacle}$.--env_dim, default=5. The environment dimension $e_d \in [3,9]$.--num_obstacles, default=2. The number of obstacles $n_o \ge 0$ in the environment.
Results
After training, a timestamped folder runs/run-<year>_<month>_<day>_<hour>_<minute>_<second>/ is created with results from the training and evaluation procedure:
* The trajectories per evaluation scene
* Metric plots
* Average cummulative reward
* Average steps
* Ending causes (e.g., collision or goal reaching)
Owner
- Name: Francisco Neves
- Login: Fsn9
- Kind: user
- Location: Porto
- Repositories: 2
- Profile: https://github.com/Fsn9
Citation (CITATION.cff)
cff-version: 1.2.0 message: "If you use this software, please cite it as below." authors: - name: Francisco Neves title: "rl2d: Learning to navigate in a 2D world using RL" version: 1.0 date-released: 2022-10-10