This is a custom environment with the OpenAI Gym interface. The Blue sphere is scripted to always go towards the Red sphere. The Red sphere is controlled through the env.step interface, possibly by a trained agent.
The goal is for Red to learn to approach its target point Yellow while avoiding Blue, and the obstacle blocks.
To test the environment:
python test.py