Skip to content

Game Environments Overview ​

This project includes 9 classic multi-agent game scenarios.

Environment Summary ​

#GameTypeAgentsChallengeCommAction Dim
1SimpleNavigation1Navigate to landmark✗5
2SpreadCooperativeN (3)Cover landmarks + avoid collisions✗5
3AdversaryAdversarial1+NDeception & inference✗5
4PushAdversarial1+1Physical blocking✗5
5TagChase3+1Cooperative hunting vs evasion✗5
6CryptoCommunication3Encryption vs eavesdropping✓4
7ReferenceCommunication2Bidirectional signaling✓15
8Speaker-ListenerCommunication1+1Unidirectional navigation✓3/5
9World CommLarge-scale4+2Leader broadcast + team hunting✓9/5

Common Physics ​

ParameterValueDescription
Time step dt0.1Per-step simulation time
Damping0.25Velocity decay: v ← 0.75 × v
Mass1.0Default agent mass
Sensitivity5.0Action-to-force multiplier
Map bounds[-1, 1]²X, 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)

Built with ❤️ for Multi-Agent AI Research