3DQ (Trix)

A Unified Kinematic Model + Explainable 3D Q-Learning Framework

One single equation → Parking (Parallel (both sides), perpendicular, angled) as well as overtaking

First paper of the 3DQ Series submitted to IEEE Transactions on Intelligent Vehicles – 2025 First-round decision expected: early 2026 Full source code will be released immediately upon acceptance of all 3DQ Series (Trix).

Key Features

• Single kinematic formulation covering all parking maneuvers + overtaking
• Guided 3D Q-Learning with hierarchical 8-cube exhaustive search
• Average convergence in under 40 episodes
• Real-time inference (< 15 ms on CPU)
• Full 3D mechanical simulation in Webots with accurate BMW i3 dynamics
• Successfully parks on both left and right sides of the road

Results (Webots R2023b)

Test_Parking_3DQ_Bagheri.mp4

Left_Parking_3DQ.mp4

Right_Parking_3DQ.mp4

* Three videos showcase real-time 3DQ (Trix) Capability

Coming Soon

• 3DQ-Overtaking (manuscript in preparation as well as Trix as a compelete work)
• Real-world deployment on full-scale autonomous vehicle (LiDAR + camera + ultrasonics)
• ROS 2 + NVIDIA Jetson implementation

Author

A. Bagheri B.Sc. Mechanical Engineering & Computer Engineering – Iran University of Science and Technology, 2026

GitHub: @aibgr | LinkedIn: @aibgr

Supervisor: Dr. Hassan Zabihi – Iran University of Science and Technology

If you find this work useful, please give it a ⭐ Full source code will be made public immediately after publication of the 3DQ(Trix) series.

More exciting updates coming very soon!