---
license: mit
task_categories:
  - reinforcement-learning
tags:
  - gymnasium
  - lunar-lander
  - imitation-learning
  - video
---

# Mind Games — Demos

Gameplay demonstrations for the [mind-games](https://github.com/kvark/mind-games) project.

## Lunar Lander

200 episodes of LunarLander-v3 (Gymnasium) collected with a heuristic PD-controller expert.

- **Format**: H.264 video (84×84) + NumPy observation arrays + JSON metadata per episode
- **Size**: ~35MB
- **Mean reward**: ~170 (expert heuristic averages ~163)
- **Structure**: `lunar_lander/episode_NNNN/` with `frames.mp4`, `observations.npy`, `metadata.json`
- **Index**: `lunar_lander/index.json` with per-episode reward and step count