Skip to content

Adversary — Deception & Inference ​

Environment

  • PettingZoo: simple_adversary_v3 | Agents: 1 Adversary + N Good (default 2)

Objective ​

  • Good Agents: Know the target; deceive Adversary by splitting (Scorer + Decoy)
  • Adversary: Doesn't know target; must infer from Good Agents' behavior

Zero-sum: Good wants Adversary far from target; Adversary wants to be close.

Key Observations ​

  • Good Agent: Has goal field (target landmark position) — unique advantage
  • Adversary: Sees all landmarks but no goal indicator

Run ​

bash
python adv_API.py

Built with ❤️ for Multi-Agent AI Research