HIGL
This is a PyTorch implementation for our paper: Landmark-Guided Subgoal Generation in Hierarchical Reinforcement Learning (NeurIPS 2021).
Our code is based on official implementation of HRAC (NeurIPS 2020) and Map-planner (NeurIPS 2019)
Installation
conda create -n higl python=3.6
conda activate higl
./install_all.sh
Also, to run the MuJoCo experiments, a license is required (see here).
Usage
Training & Evaluation
- Point Maze
./scripts/point_maze_sparse.sh ${reward_shaping} ${timesteps} ${gpu} ${seed}
./scripts/point_maze_sparse.sh dense 5e5 0 2
./scripts/point_maze_sparse.sh sparse 5e5 0 2
- Ant Maze (U-shape)
./scripts/higl_ant_maze_u.sh ${reward_shaping} ${timesteps} ${gpu} ${seed}
./scripts/higl_ant_maze_u.sh dense 10e5 0 2
./scripts/higl_ant_maze_u.sh sparse 10e5 0 2
- Ant Maze (W-shape)
./scripts/higl_ant_maze_w.sh ${reward_shaping} ${timesteps} ${gpu} ${seed}
./scripts/higl_ant_maze_w.sh dense 10e5 0 2
./scripts/higl_ant_maze_w.sh sparse 10e5 0 2
- Reacher & Pusher
./scripts/higl_fetch.sh ${env} ${timesteps} ${gpu} ${seed}
./scripts/higl_fetch.sh Reacher3D-v0 5e5 0 2
./scripts/higl_fetch.sh Pusher-v0 10e5 0 2
- Stochastic Ant Maze (U-shape)
./scripts/higl_ant_maze_u_stoch.sh ${reward_shaping} ${timesteps} ${gpu} ${seed}
./scripts/higl_ant_maze_u_stoch.sh dense 10e5 0 2
./scripts/higl_ant_maze_u_stoch.sh sparse 10e5 0 2
4 Oct 08, 2022
92 Jan 04, 2023
303 Mar 15, 2022
12 Jan 10, 2022
68 Jan 02, 2023
165 Nov 04, 2022
270 Dec 30, 2022
5k Jan 04, 2023
29 Dec 29, 2022
83 Dec 14, 2022
57 Dec 29, 2022
204 Dec 25, 2022
0 Dec 09, 2021
110 Nov 21, 2022
21 Dec 19, 2022
112 Jan 04, 2023
4 Dec 11, 2022
14 Nov 10, 2022
14 Dec 17, 2022