This repository contains code accompaning the paper, Divide-and-Conquer Reinforcement Learning (Ghosh et al., ICLR 2018). It includes code for the DnC algorithm, and the Mujoco environments used for the empirical evaluation. Please see the project website for videos and further details.
This codebase requires a valid installation of rllab. Please refer to the rllab repository for installation instructions.
The environments are built in Mujoco 1.31: follow the instructions here to install Mujoco 1.31 if not already done. You are required to have a Mujoco license to run any of the environments.
Sample scripts for working with DnC and the provided environments can be found in the examples directory. In particular, a sample scripts for running DnC is located here.
source activate rllab_env
python examples/dnc_pick.pyEnvironments are located in the dnc/envs/ directory, and the DnC implementation can be found at dnc/algos/.
To ask questions or report issues, please open an issue on the issues tracker.
If you use DnC, please cite the following paper:
- Dibya Ghosh, Avi Singh, Aravind Rajeswaran, Vikash Kumar, Sergey Levine. "Divide-and-Conquer Reinforcement Learning". Proceedings of the International Conference on Learning Representaions (ICLR), 2018.