Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AI-Olympics: Exploring the Generalization of Agents through Open Competitions

Chen Wang, Yan Song, Shuai Wu, Sa Wu, Ruizhi Zhang, Shu Lin, Haifeng Zhang

TL;DR

The paper introduces AI-Olympics, a Python-based 2D physics engine with continuous control, partial observability, and zero-sum, multi-task scenarios to study generalization in multi-agent decision-making. It details the AI-Olympics environment, a suite of six scenarios plus integrated games, and an online competition series on the Jidi platform with Swiss-system pairings and real-time rankings to evaluate map, scenario, and opponent generalization. Key findings show that broader training data improves map generalization, scenario-specific strategies enhance performance, and aggressive behaviors can emerge as advantageous tactics, with diverse AI methods powering practical engineering insights. The work provides a scalable, community-driven platform to advance understanding of generalization in complex, adversarial, multi-task environments and suggests directions for expanding tasks and evaluation practices.

Abstract

Between 2021 and 2023, AI-Olympics, a series of online AI competitions was hosted by the online evaluation platform Jidi in collaboration with the IJCAI committee. In these competitions, an agent is required to accomplish diverse sports tasks in a two-dimensional continuous world, while competing against an opponent. This paper provides a brief overview of the competition series and highlights notable findings. We aim to contribute insights to the field of multi-agent decision-making and explore the generalization of agents through engineering efforts.

AI-Olympics: Exploring the Generalization of Agents through Open Competitions

TL;DR

The paper introduces AI-Olympics, a Python-based 2D physics engine with continuous control, partial observability, and zero-sum, multi-task scenarios to study generalization in multi-agent decision-making. It details the AI-Olympics environment, a suite of six scenarios plus integrated games, and an online competition series on the Jidi platform with Swiss-system pairings and real-time rankings to evaluate map, scenario, and opponent generalization. Key findings show that broader training data improves map generalization, scenario-specific strategies enhance performance, and aggressive behaviors can emerge as advantageous tactics, with diverse AI methods powering practical engineering insights. The work provides a scalable, community-driven platform to advance understanding of generalization in complex, adversarial, multi-task environments and suggests directions for expanding tasks and evaluation practices.

Abstract

Between 2021 and 2023, AI-Olympics, a series of online AI competitions was hosted by the online evaluation platform Jidi in collaboration with the IJCAI committee. In these competitions, an agent is required to accomplish diverse sports tasks in a two-dimensional continuous world, while competing against an opponent. This paper provides a brief overview of the competition series and highlights notable findings. We aim to contribute insights to the field of multi-agent decision-making and explore the generalization of agents through engineering efforts.
Paper Structure (20 sections, 3 figures, 1 table)

This paper contains 20 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. AI-Olympics Environment
  3. Mobility and Observability
  4. Zero-sum Games
  5. Multiple Designed Tasks
  6. Running game.
  7. Other single games.
  8. Integrated games.
  9. AI-Olympics Competition Series
  10. Evaluation Protocol
  11. Generalization Abilities Assessment
  12. Map generalization.
  13. Scenario generalization.
  14. Opponent generalization.
  15. Analysis of the Submitted Agents
  16. ...and 5 more sections

Figures (3)

  • Figure 1: An illustration of agent's mobility and vision. An agent consumes stored energy and moves, while only observing partially of its surroundings.
  • Figure 2: Three example maps in Running game. Agents are represented as circles and the arrow attached to them shows the driving force applied.
  • Figure 3: Five additional scenarios built from AI-Olympics game engine.