This repository is a codebase for Vector Policy Fields using Reinforcement Learning Networks. While the framework primarily allows for learning control policies in continuous state spaces and action spaces, the modules in this repository afford cool solutions to a variety of RL problems, including:
- Learning control policies in continuous state and action spaces from demonstrations.
- Deep Inverse Reinforcement Learning in 3D! (Please stay tuned for 4 dimensions).
- Learning transition dynamics of MDPs as convolutional filters.
The code in this repository builds off of the paper Reinforcement Learning via Recurrent Convolutional Neural Networks, and the associated repository https://github.com/tanmayshankar/RCNN_MDP.