Work

Vision

Deep RL & Hugging Face

Doom and a walking ant are how I like to see whether an agent has learned anything. These are PPO and DQN exercises from the Hugging Face Deep RL course, in ViZDoom and PyBullet.

A PPO agent playing Doom: the view swings across a canyon while the HUD tracks health and ammunition.
Fig. 1The PPO agent in ViZDoom, eight seconds of one rollout. One recorded rollout from a course exercise in ViZDoom.
A PyBullet Ant robot with four coloured legs walking across a blue and white checkered floor.
Fig. 2The PyBullet Ant, five seconds of one rollout. One recorded rollout from a course exercise, cropped and slowed to 8 fps.
Stack
Deep RL, PPO, DQN, Hugging Face
Links
Code

Also