payamn/d4pg-pytorch

PyTorch implementation of Distributed Distributional Deterministic Policy Gradients (https://arxiv.org/abs/1804.08617)

D4PG-pytorch

PyTorch implementation of Distributed Distributional Deterministic Policy Gradients (https://arxiv.org/abs/1804.08617). Supported environments

• Pendulum-v0
• LunarLanderContinous-v2
• BipedalWalker-v2

Usage

Run train.py to run experiment specified in config.yaml.

Tests

In progress, for now tests can be used for reproducing results.

Demo

Detailed results of training can be found at

• Pendulum-v0
  • d3pg, d3pg prioritized
  • d4pg, d4pg prioritized
• LunarLanderContinuous-v2
  • d3pg, d3pg prioritized
  • d4pg, d4pg prioritized
• BipedalWalker-v2
  • d3pg, d3pg prioritized
  • d4pg, d4pg prioritized

Acknowledgements

The project partly based on the Mark Sinton TensorFlow implementation, which helped greatly in difficult parts.