gsartoretti/PRIMAL
gsartoretti/PRIMAL has 449 GitHub stars and 78 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-17 19:50 UTC. Last push 2024-04-09.
Measured star growth
- in 24h
- 0
- in 7 days
- Collecting history
- in 15 days
- Collecting history
- in 1 month
- Collecting history
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
About
PRIMAL: Pathfinding via Reinforcement and Imitation Multi-Agent Learning -- Distributed RL/IL code for Multi-Agent Path Finding (MAPF)
LearningREADME.md
OriginalPRIMAL: Pathfinding via Reinforcement and Imitation Multi-Agent Learning
Reinforcement learning code to train multiple agents to collaboratively plan their paths in a 2D grid world, as well as to test/visualize the learned policy on handcrafted scenarios.
NEW: Please try the brand new online interactive demo of our trained PRIMAL model! You can customize the grid size, add/remove obstacle, add agents and assign them goals, and finally run the model online and see the results.
File list
- DRLMAPF_A3C_RNN.ipynb: Multi-agent training code. Training runs on GPU by default, change line "with tf.device("/gpu:0"):" to "with tf.device("/cpu:0"):" to train on CPU (much slower).
- mapf_gym.py: Multi-agent path planning gym environment, in which agents learn collective path planning.
- primal_testing.py: Code to run systematic validation tests of PRIMAL, pulled from the saved_environments folder as .npy files (examples available here) and output results in a given folder (by default: primal_results).
- mapf_gym_cap.py: Multi-agent path planning gym environment, with capped goal distance state value for validation in larger environments.
- mapgenerator.py: Script for creating custom environments and testing a trained model on them. As an example, the trained model used in our paper can be found here.
Before compilation: compile cpp_mstar code
- cd into the od_mstar3 folder.
- python3 setup.py build_ext (may need --inplace as extra argument).
- copy so object from build/lib.*/ at the root of the od_mstar3 folder.
- Check by going back to the root of the git folder, running python3 and "import cpp_mstar"
Custom testing
Edit mapgenerator.py to the correct path for the model. By default, the model is loaded from the model_primal folder.
Hotkeys:
- o: obstacle mode
- a: agent mod
- g: goal mode, click an agent then click a free tile to place its goal
- c: clear agents
- r: reset
- up/down arrows: change size
- p: pause inference
Requirements
- Python 3.4
- Cython 0.28.4
- OpenAI Gym 0.9.4
- Tensorflow 1.3.1
- Numpy 1.13.3
- matplotlib
- imageio (for GIFs creation)
- tk
- networkx (if using od_mstar.py and not the C++ version)