AutoCarSim v0.0.1

Urban Autonomous Vehicle Simulation – Release 0.0.1

This is the first release of the project, marking a transition from early training experiments to a working prototype in a simulated urban setting. It represents the last state of the project as of June 2023, reworked with a simple UI so that it can be built into a packaged release.

🔹 What’s included

• Vehicle physics simulation in Unity, including chassis mechanics, Ackermann-Jeantaud steering, suspension modeling, engine torque curves, transmission, and automatic gearbox.

• Urban environment generation from OpenStreetMap data, with triangulation for rendering buildings, roads, and structures in 3D.

• AI model trained in the second environment (basic navigation + obstacle avoidance) directly deployed into the third environment (the city).

  • The model relies on ray-based obstacle detection and a reward system encouraging safe navigation while moving toward goals.
  • It is not yet trained specifically for the complexity of the city but demonstrates baseline transfer from simpler environments.

• Pathfinding with A* implemented to provide the AI with city navigation steps.

🔹 Key observations

• The agent is able to move through the city and avoid obstacles but sometimes:

  • Hugs the right side of the road excessively,
  • Misinterprets bridges as unreachable targets,
  • Prefers taking longer paths around buildings instead of following roads directly.

🔹 Next steps

• Extend training directly in the city environment.
• Encourage road-following behavior and improve handling of bridges and complex structures.
• Refine sensor sensitivity and reward shaping for more natural driving.

This release demonstrates the first integration of physics, environment modeling, and AI into a single working demo, laying the foundation for potential future iterations.