Distributed Assembly

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.

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)

Authors

Guillaume Sartoretti

Justin Kerr