Scheduled Policy Optimization for Natural Language Communication with Intelligent Agents

Models and Algorithms

See files under walk_the_blocks/BlockWorldRoboticAgent/srcs/

• learn_by_ppo.py run this file for training, you can change the schedule mechanism in the function ppo_update(), these are the options:

  • do imitation every 50
  • do imitation based on rules
  • imitation 1 epoch and then RL 1 epoch

  example: python learn_by_ppo.py -lr 0.0001 -max_epochs 2 -entropy_coef 0.05

• policy_model.py the network achitecture and loss functions:

  • PPO Loss
  • Supervised Loss
  • Advantage Actor-Critic Loss

Instructions

For the usage of the Block-world environment, please refer to https://github.com/clic-lab/blocks

Train the RL agents

• S-REIN *

If you use our code in your own research, please cite the following paper

@article{xiong2018scheduled,
  title={Scheduled Policy Optimization for Natural Language Communication with Intelligent Agents},
  author={Xiong, Wenhan and Guo, Xiaoxiao and Yu, Mo and Chang, Shiyu and Zhou, Bowen and Wang, William Yang},
  journal={arXiv preprint arXiv:1806.06187},
  year={2018}
}