Sparse-Reward-Algorithms

Introduction

We implemented many classes of Sparse Reward algorithms in Gym Fetch environment including Reward Shaping, Imitation Learning, Curriculum Learning, Hindsight Experience Replay, Curiosity-Driven Exploration, Hierachical Reinforcement Learning. This work is for better understanding of sparse reward algorithms.

Our code is based on https://github.com/andrew-j-levy/Hierarchical-Actor-Critc-HAC- and we have changed a lot on code simplification and content richness.

Usage

1. DDPG:

python main.py --retrain

2. Reward Shaping:

python main.py --retrain --rtype dense

3. Curriculum Learning:

python main.py --retrain --curriculum 2

4. Imitation Learning:

python main.py --retrain --imitation --imit_ratio 1

5. Hindsight Experience Replay:

python main.py --retrain --her

6. Forward Dynamic:

python main.py --retrain --curiosity

7. Hierachical DDPG:

python main.py --retrain --layers 2

8. Test the latest saved checkpoint:

python main.py --test

if using HDDPG, you should use :

python main.py --test --layers 2

9. Save demostrations for imitation learning：

python main.py --retrain --her --save_experience

Result