Strategic Coopetition

Computational techniques for modelling strategic coopetition (the simultaneous pursuit of cooperation and competition) in mixed-motive multi-agent environments. Bridges conceptual modelling, computational game theory, and reinforcement learning.

Trust × cooperation phase space with Nash equilibrium (square) and Pareto-optimal point (star). Four colour-coded behavioural regions: sustainable cooperation, paradoxical, defection spiral, and exploitative.

• 20 multi-agent environments spanning four mechanism classes (interdependence, trust, collective action, reciprocity).
• 126-algorithm reference suite: 16 training algorithms, 7 game-theoretic oracles, 2 heuristics, and 101 constant-action policies.
• Four calibrated case studies grounded in real-world coopetitive relationships: Samsung–Sony LCD (96.7%), Renault–Nissan (81.7%), Apache HTTP Server (86.7%), Apple iOS App Store (87.3%).
• Reward-type ablation methodology for mixed-motive evaluation, varying reward mutuality across private, integrated, and cooperative configurations while holding mechanism rules fixed.
• Three-API design: Gymnasium (single-agent style), PettingZoo Parallel (simultaneous moves), and PettingZoo AEC (sequential moves).

Algorithm reference suite organised by paradigm. The benchmark covers four learning families plus heuristic and game-theoretic oracle baselines.

git clone https://github.com/vikpant/strategic-coopetition.git
cd strategic-coopetition/coopetition_gym
pip install -e .

import coopetition_gym

env = coopetition_gym.make("TrustDilemma-v0")
obs, info = env.reset(seed=42)

for _ in range(100):
    obs, reward, terminated, truncated, info = env.step([60.0, 55.0])
    if terminated or truncated:
        break

A runnable Jupyter walkthrough lives at coopetition_gym/examples/quickstart.ipynb.

The hosted documentation site is at https://vikpant.github.io/strategic-coopetition/ and is built automatically from coopetition_gym/docs/ on every push to master by the pages.yml workflow. The site covers installation, the API reference, the 20 environments, the evaluation protocol, the four mechanism-class theory chapters, tutorials, and troubleshooting.

If you use Coopetition-Gym in your research, please cite the relevant technical report.

@article{pant2025interdependence,
  title   = {Computational Foundations for Strategic Coopetition: Formalizing Interdependence and Complementarity},
  author  = {Pant, Vik and Yu, Eric},
  journal = {arXiv preprint arXiv:2510.18802},
  year    = {2025}
}

@article{pant2025trust,
  title   = {Computational Foundations for Strategic Coopetition: Formalizing Trust and Reputation Dynamics},
  author  = {Pant, Vik and Yu, Eric},
  journal = {arXiv preprint arXiv:2510.24909},
  year    = {2025}
}

@article{pant2026collective,
  title   = {Computational Foundations for Strategic Coopetition: Formalizing Collective Action and Loyalty},
  author  = {Pant, Vik and Yu, Eric},
  journal = {arXiv preprint arXiv:2601.16237},
  year    = {2026}
}

@article{pant2026reciprocity,
  title   = {Computational Foundations for Strategic Coopetition: Formalizing Sequential Interaction and Reciprocity},
  author  = {Pant, Vik and Yu, Eric},
  journal = {arXiv preprint arXiv:2604.01240},
  year    = {2026}
}

@article{pant2026coopetitiongym,
  title   = {Coopetition-Gym v1: A Formally Grounded Platform for Mixed-Motive Multi-Agent Reinforcement Learning under Strategic Coopetition},
  author  = {Pant, Vik and Yu, Eric},
  journal = {arXiv preprint arXiv:2605.02063},
  year    = {2026}
}

Questions, ideas, and proposals are welcome on the project's GitHub Discussions board. Bug reports and feature requests should be filed via GitHub Issues.

Contributions are welcome. See CONTRIBUTING.md and the Code of Conduct.

MIT, see LICENSE.

Vik Pant, PhD (LinkedIn · Google Scholar) · Eric Yu, PhD · Faculty of Information and Department of Computer Science, University of Toronto.