Environments
The RLForge environments provide a diverse set of testbeds for experimenting with reinforcement learning agents. They are designed to illustrate different challenges such as exploration, control, and navigation, while remaining lightweight and easy to integrate with your agents.
In addition to these built-in environments, RLForge is fully compatible with all Gymnasium environments, allowing you to extend experiments to a wide variety of standardized benchmarks.
Bandits — a simple environment family for testing exploration strategies and reward maximization in multi-armed bandit problems.
Dyna Maze — a grid-based maze environment ideal for testing planning and model-based reinforcement learning algorithms.
Shortcut Maze — a variation of the maze environment with multiple paths, emphasizing exploration and the discovery of efficient trajectories.
Short Corridor — a small episodic environment with a biased action space, commonly used to illustrate policy gradient behavior and the importance of exploration in constrained settings.
Pendulum — a classic continuous control task where the agent learns to balance and swing up a pendulum.
Mecanum Car — a robotics-inspired environment simulating a mecanum-wheeled vehicle, useful for testing control in multi-dimensional continuous spaces.
Obstacle Avoidance — challenges agents to navigate safely through an environment filled with obstacles, highlighting spatial awareness and reactive control.
Trajectory Tracking — focuses on following predefined paths or trajectories, testing precision and stability in continuous control tasks.
These environments serve as practical benchmarks for evaluating different agent architectures, from tabular methods to deep reinforcement learning approaches.