This repo contains code for training a SSB64 ai agent using behavioral cloning (supervised, imitation learning), which includes:

1. Collecting behavioral cloning data (image-action trajectories) from gameplay
2. Formatting those trajectories as training examples (in various action formats and for use with feed-forward and recurrent networks)
3. Training models on the formatted datasets (feed-forward and recurrent)
4. Deploying the trained models to play (only against the game ai implemented for now)

The network controls blue DK in the following clip (see here for the full match):