lcdbezerra/phyrl

PhyRL: Physics-Informed Model-Based Reinforcement Learning

PhyRL: Physics-Informed Model-Based Reinforcement Learning

Usage:

• Train baseline, Dreamer v2
  • Get latest Dreamer v2 code: https://github.com/danijar/dreamerv2
  • Run python lunarlander_train.py on Dreamer v2 repository
• PhyRL: Train Model
  • Go to examples/lunarlander
  • Run python phyrl.py [Physics Constraint Index] where the index ranges from 1 to 4 as presented in the paper report.
  • Observe the results on Weights & Biases: https://wandb.ai/lucascamara/PhyRL
• PhyRL: Train RL Agent
  • Go to examples/lunarlander
  • Run python phyrl_agent.py [Physics Constraint Index] where the index ranges from 1 to 4 as presented in the paper report.
    • Note: The RL agent will randomly pick one learned model from Weights & Biases corresponding to the desired index, and then train on it.
  • Observe the results on Weights & Biases: https://wandb.ai/lucascamara/PhyRL_Agents

Packages:

python
numpy
scipy
pysindy
scikit-learn
torch
tqdm
pandas