The Swimmer task is a good example to test TRPO. This task involves a 3-link swimming robot in a viscous fluid, where the goal is to make it swim forward as fast as possible by actuating the two joints (http://gym.openai.com/envs/Swimmer-v2/). The following screenshot shows how Swimmer looks in the MuJoCo simulator:
To train an agent for Swimmer, run the following command under the src folder:
CUDA_VISIBLE_DEVICES= python train.py -t Swimmer
There are two arguments in train.py. One is -t, or --task, indicating the name of the MuJoCo or classic control task you want to test. Since the state spaces of these control ...