Skip to content

jaromiru/NASimEmu-agents

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NASimEmu-agents

This is a repository containing deep RL agents for NASimEmu.

Related workshop paper, and dissertation (chapter 6).

Usage

Install NASimEmu and run training as:

python main.py <path-to-scenario>

Alternatively, you can test a trained model as an example below:

mkdir out/
python main.py -load_model trained_models/mlp.pt --trace ../NASimEmu-public/scenarios/uni.v2.yaml -device cpu -net_class NASimNetMLP -use_a_t -episode_step_limit 100 -augment_with_action

Experiments excerpt

Generalization of invariant architectures vs. MLP:

Training to stop:

Last action embedding:

Comparison of architecture variants:

Scaling experiment:

main.py huge-gen-rgoal-stoch -device cpu -cpus 2 -epoch 100 -max_epochs 200 --no_debug -net_class NASimNetInvMAct -force_continue_epochs 100 -use_a_t -episode_step_limit 200 -augment_with_action

Transfer to emulation

Simulation-trained agents can be transferred to emulation, see the emulation log.

About

Deep RL agents for NASimEmu. See also https://github.com/jaromiru/NASimEmu.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages