Change the blending flag from BLEND_RGBA_MULT to BLEND_RGB_MULT to fix the rendering issue in draw function. This ensures proper display of elements on the screen.

Adapted gym env to only del a visualizer if the environment name is in the dictionary, as this is not the case during intialization of the rl env.

Included short description on how to use the reinforcement learning and how to convert overcooked-ai layouts.

Switched from using np.round to np.ceil for block size calculation to ensure more accurate grid sizes. Additionally, replaced range with np.arange for consistent step intervals, and ensured rectangle positions use np.round for precise placement.

Replaced grid_size-based scaling with a fixed value of 0.15 for the polygon creation offset. This change ensures consistency in offset scaling regardless of the grid size.

Set the environment's end time based on iteration count for performance benchmarking. Also, increase the number of iterations from 500 to 1,000 for more robust testing results.

Updated `gym_env.py` and `drawing.py` to use unique environment names (`env_name`) for caching purposes, leveraging UUIDs for better cache management. Revised render and image fetch methods to incorporate the unique environment identifier, ensuring accurate and efficient cache invalidation and retrieval.

Enhanced visualizer by adding cache flags for background and counters and a reduced background mode for optimization. Adjusted relevant methods across video replay, performance benchmark, gym environment, and GUI code to use the new caching logic.

Deleted obsolete drawing_perf_benchmark.py and replaced it with the more versatile performance_benchmark.py. Consolidated individual config file reads into a unified load_config_files function, which is now used across various modules for increased code readability and maintainability.

Enhanced `get_state_image` to accept a list of controlled players, defaulting to a single player if not provided. This change ensures that the drawing function correctly interprets the input for controlled players, improving the function's flexibility and usability.

Renamed `grid_size` to `node_size` for clarity and consistency. Removed commented-out code and redundant parameters, and optimized some conditional statements to improve readability.

Adapted LayoutConverter to not be integrated in gym_env.py but to be called seperately with a command line argument. Additionally, all layouts are now stored in the config/layouts folder and no layout files with a different format are stored within our project.

Changed layout converter path management to be callable outside of the main program with the current folder structure.

Transferring the layouts and creating the configs. Additionally, enabling the choice of random or determinstic order generation in the configs.

Inference uses the same configs as the training and instantiates the model load function from the configs.

Removed unnecessary debug print statements from environment and orders modules. Refined test cases in test_order_manager.py to add clarity and coverage without redundant checks.

Reverted the default study configuration from `study_config_deterministic_order.yaml` back to `study_config.yaml`. This ensures that the standard configuration is used by default as opposed to the deterministic order configuration.