This project is my own implementation of TD3 paper for solving continuous control problem in the Bipedal Walker environment of OpenAI Gym.
Twin Delayed Deep Deterministic Policy Gradient (TD3) is an Actor Critic reinforcement learning algorithm. This algorithm is an improved version of Deep Deterministic Policy Gradient. It uses 2 critic networks to limit the overestimation bias. Further, it uses delayed policy and target networks updates to reduce per-update error.
This project requires following dependencies:
- Python
- Tensorflow
- Numpy
- OpenAI Gym
- Jupyter Notebook (optional)
For trying out, download the weights from repository and put them in the same folder. Run all cells except the Training cell with comment #Training
on top.