Game Environments Overview
This project includes 9 classic multi-agent game scenarios.
Environment Summary
| # | Game | Type | Agents | Challenge | Comm | Action Dim |
|---|---|---|---|---|---|---|
| 1 | Simple | Navigation | 1 | Navigate to landmark | ✗ | 5 |
| 2 | Spread | Cooperative | N (3) | Cover landmarks + avoid collisions | ✗ | 5 |
| 3 | Adversary | Adversarial | 1+N | Deception & inference | ✗ | 5 |
| 4 | Push | Adversarial | 1+1 | Physical blocking | ✗ | 5 |
| 5 | Tag | Chase | 3+1 | Cooperative hunting vs evasion | ✗ | 5 |
| 6 | Crypto | Communication | 3 | Encryption vs eavesdropping | ✓ | 4 |
| 7 | Reference | Communication | 2 | Bidirectional signaling | ✓ | 15 |
| 8 | Speaker-Listener | Communication | 1+1 | Unidirectional navigation | ✓ | 3/5 |
| 9 | World Comm | Large-scale | 4+2 | Leader broadcast + team hunting | ✓ | 9/5 |
Common Physics
| Parameter | Value | Description |
|---|---|---|
Time step dt | 0.1 | Per-step simulation time |
| Damping | 0.25 | Velocity decay: v ← 0.75 × v |
| Mass | 1.0 | Default agent mass |
| Sensitivity | 5.0 | Action-to-force multiplier |
| Map bounds | [-1, 1]² | X, Y range |
Common Action Space
5D continuous: [no_op, left, right, down, up] ∈ [0,1]⁵
Net force: F_x = (right - left) × sensitivity, F_y = (up - down) × sensitivity
Important
- Never activate opposing directions simultaneously
- Reduce thrust when approaching target
- Thrust toward center when near boundaries (|x| or |y| > 0.9)