Environment
The competition adopts multiple scenarios built on the AI-Olympics engine, a 2D Pythonic physical simulator created from scratch with simple dependencies. In these scenarios:
- Each agent/player is simulated as an elastic ball that can collide with other agents/players or the objects in the map.
- The agent has limited energy and applying the driving force consumes energy at a particular rate. The player will lose control of the agents when running out of energy.
- The player has partial observation limited to a rectangular area around the player itself.
- At each discretized time-step, the agent makes decisions on both the magnitude of driving forces and turning angles to control the movement.
- The color of objects in the map represents its attribute and needs to be recognised.
The IJCAI 2023 AI-Olympics Competition chooses six game scenarios. In each scenario:
- Two agents compete for the same goal;
- In the scenario of running, the goal is to reach the final as fast as possible;
- In the scenario of table-hockey, the goal is to hit the ball into opponent's goal through collision and to defend their own goal.
- In the scenario of football, the agent shares the same goal as in table-hockey, except that they can now move freely on the map.
- In the scenario of wrestling, the goal is to push the opponent out of bounds while staying in bounds.
- In the scenario of billiards, the goal is to score as many as possible by hitting the ball of their own color into the basket.
- In the scenario of curling, the goal is to throw the rock as close to the center as possible.
- Winner of each scenario receives +1 score and the player with a higher score after completing all scenarios wins the game.
