Skip to content

Latest commit

 

History

History
18 lines (14 loc) · 1.18 KB

README.md

File metadata and controls

18 lines (14 loc) · 1.18 KB

Deep RL Projects

Implementation of deep reinforcement learning models

Models

  • Soft Actor Critic
  • DARC
  • GAIL (In Progress)

SAC

Soft actor critic is an off-policy model that attempts to maximize reward as well as entropy of its actions. With its objective being

This pushes the policy to balance between exploration and exploitation of its environment with minimum number of hyperparameters to tune.
The policy uses a gaussian distribution for continuous action prediction and the value network uses a twin q-net to prevent explosive growth in reward.

DARC

DARC builds on top of SAC for transfer from source to target domain by attempting to match transition probabilities. This is done through an additional classifier for classification between source and target domains and adding reward based on dynamics adaptation.