zanghyu / TMAP

TMAP: Integrating Trust Region and Maximum Entropy with Augmented Bellman Equation for Policy Optimization

TMAP

Code for the paper TMAP: Integrating Trust Region and Maximum Entropy with Augmented Bellman Equation for Policy Optimization

This code is based on openai spinning up

Requirements

We assume you have access to a gpu that can run CUDA. All of the dependencies are in the setup.py file.

pip install -e .

After the installation ends you can run the code.

Instructions

To train a TMAP agent on the Ant-v3 task, you can first change your path to the_path_of_TMAP/spinup/algo/tmap/ and then run the following code

CUDA_VISIBLE_DEVICES=0 python tmap.py --md 0.08 --epoch 750 --env Ant-v3