How jtidbury2/thirdyearproject-autonomousintersectionmanagment works
• JavaSim – the original AIM simulator, used as a blueprint (even though it’s no longer used directly).
• PettingZooSim – a PettingZoo-based environment specifically set up to train deep reinforcement learning agents on highway and intersection scenarios.
• PythonSim – an extended intersection simulation (derived from an open–source simulator) that seems to provide the main engine for running the research simulations.
a. The “Blueprint/Reference Simulator” component using JavaSim.
b. The “Deep RL Simulation Environment” via PettingZooSim, which itself is subdivided into modules such as:
– HighwayEnv (with its environments, vehicle dynamics, road models, etc.).
– Documentation and test suites (helpful to understand simulation parameters).
c. The “Research Simulator and UI” through PythonSim, which includes the simulation engine, settings (lib/settings.py), visualization (my_main_window.py, my_paint_canvas.py), logging, and execution components.
d. The “Reinforcement Learning Agents” component from rl-agents. This part shows multiple agent types (DQN, MCTS, budgeted_ftq, etc.) with configurations for various environments provided via a rich set of JSON configuration files.
Interactions and data flows
– The research workflow appears to start by configuring the RL agents (using rl-agents and helper scripts) and then training/validating them using the environments provided by PettingZooSim.
– Simulation results and evaluation logs are generated by both PettingZooSim (with numerous “scripts/” and “tests” folders) and PythonSim (which contains experiment run files, logging, and visualization scripts).
Architectural patterns and design principles
– There is a clear separation of concerns: simulation engine (PythonSim), environment definition and agent training (PettingZooSim with its “highway_env” and multi-agent configurations), and agent definitions (rl-agents).
– The design also reflects modularity. For example, runtime configurations and experiment scripts for RL experiments are separately organized (notice the multiple “configs” directories under rl-agents/scripts) ensuring that different strategies or environments can be plugged into the simulation.
– Testing and documentation are an integral part (with numerous tests and docs directories), indicating a robust development process.
Relevant technologies and libraries
– For PettingZooSim, the underlying PettingZoo library is the basis.
– In rl-agents, deep RL implementations are provided in frameworks like PyTorch (as seen in files such as pytorch.py).
– PythonSim closely overlaps with traditional simulation software design – incorporating GUI elements and logging, likely using Python’s standard libraries along with custom modules (e.g., my_main_window, my_paint_canvas).
– Build and configuration files (like setup.py, pyproject.toml, Makefile in docs) underline that package management and documentation generation are part of the design.