Simulation Tags

Use tags to quickly find simulations suited to your algorithm or learning goal:

Competitive

The simulation pits two or more agents against each other. Only one agent (or team) can win — cooperation is not the goal. This is a subset of Multi-Agent.

Cooperative

Two or more agents must work together to achieve a shared goal. Agents are rewarded collectively, not against each other. This is a subset of Multi-Agent.

Dense Reward

The simulation returns a meaningful reward signal after every step, not just at episode end. This makes it well-suited to gradient-based RL methods such as PPO or Q-learning.

Deterministic

The simulation contains no random elements. Given the same state, the same action always produces the same next state. Planning and search algorithms (e.g. minimax, MCTS) are directly applicable.

Experimental

The simulation is not final and likely to change. Often this is due to balancing issues where the final scoring system has not been agreed.

Don’t expect results to be consistent between releases.

Image

The simulation sends an image to the agents rather than a well-structured data object.

Multi-Agent

The simulation requires more than one connected agent to run.

Partial Observation

The agent cannot see the full game state. Some information is hidden (e.g. mine positions, opponent cards). Algorithms must reason under uncertainty rather than planning over a known state space.

Perfect Information

The full game state is visible to all agents at every step. Planning and search algorithms can exploit this directly.

Random

The simulation includes random elements — initial conditions, environmental transitions, or both.

Single-Agent

The simulation is designed to run with exactly one agent.

Sparse Reward

The simulation only signals success or failure at the end of an episode, with little or no reward during play. This makes credit assignment harder; shaping or curriculum approaches may be needed.