yyystg / alwayssafe

Code for the paper "AlwaysSafe: Reinforcement Learning Without Safety Constraint Violations During Training"

AlwaysSafe

Code for the paper "AlwaysSafe: Reinforcement Learning Without Safety Constraint Violations During Training" — Thiago D. Simão, Nils Jansen and Matthijs T. J. Spaan, published at AAMAS 2021.

[Details]

modules

• agents: model based RL agents that interact with the environment.
• planners: the planners used by the RL agents to compute the policy in each episode.
• scripts: each file is related to one of the experiments from the paper.
• tests: mostly unittest scripts.
• util: contains common scripts to train an RL agent and evaluate a policy.

lp solver

By default, the code uses gurobipy if found, otherwise it uses cvxpy.

usage

1. install dependencies

   pipenv install
   

2. run tests

   pipenv run python -m unittest

3. reproduce the experiments

   pipenv run python -m scripts.simple
   pipenv run python -m scripts.factored
   pipenv run python -m scripts.cliff_walking

citing

@inproceedings{Simao2021alwayssafe,
  author    = {Sim{\~a}o, Thiago D. and Jansen, Nils and Spaan, Matthijs T. J.},
  title     = {AlwaysSafe: Reinforcement Learning Without Safety Constraint Violations During Training},
  year      = {2021},
  booktitle = {Proceedings of the 20th International Conference on Autonomous Agents and MultiAgent Systems (AAMAS)},
  publisher = {IFAAMAS},
  pages     = {1226–1235},
}