A PyTorch Implementation of PlaNet: A Deep Planning Network for Reinforcement Learning [1] by Danijar Hafner et.al.
- Run
main.pyfor training. - Run
eval.pyfor evaluation of a saved checkpoint. - Tensorboard will be used to display and store metrics and can be viewed by running the following:
$ tensorboard --logdir <path_to_repository>/results- Visit tensorboard in your browser! By default tensorboard launches at
localhost:6006. You might see a screen similar to this:
The video on the left is the downscaled version of the gym render.
The one on the right is generated by the decoder model.
Install dependencies ...
pytorch==1.4.0tensorboard-pytorch==0.7.1tqdm==4.42.1torchvision==0.5.0gym==0.16.0