Research into Vehicle Traversal (2024)

Summary

Role: -

Team size: 1

Time span: 6 months

Tools:

Unity

Links:

Thesis Paper

Thesis Archive

During my employment as a programmer in Saab, a comprehensive solution for dynamic vehicle traversal in simulation environment was needed. Because the performance of the simulation was critical, use of traditional motion planning algorithms was unfeasible. Instead, I searched for a solution based on NavMesh. I decided to dive into this research topic as a part of my Bachelor's thesis.

Autonomous navigation for vehicles presents unique challenges due to turning constraints and movement limitations. While humanoid agents can rotate freely, vehicles require planned movement to maintain realism. This thesis addresses this problem by proposing two solutions that balance natural movement with computational efficiency.

The first approach ensures reliable path generation within a NavMesh, allowing vehicles to follow optimized trajectories. The second method prioritizes the vehicle’s ability to navigate dynamically, adjusting its movement strategy based on environmental constraints. Both methods are compared to Hybrid A*, a well-established motion planning algorithm, with a focus on both objective efficiency and subjective user perception.

Key Features

🚗 Navigation in Virtual 3D Environments – Developing movement systems for autonomous vehicles within performance-constrained environments.
🗺️ NavMesh-based Pathfinding – Utilizing navigation meshes as the foundation for vehicle movement rather than complex motion planning algorithms.
🔀 Two Distinct Approaches:
- 🛤️ Path-focused: Creating reliable, optimized routes for vehicles to follow.
- 🚦 Vehicle-focused: Designing movement mechanics that adapt to the environment naturally.
📊 Comparative Evaluation – Implementing and testing against Hybrid A*, a state-of-art motion planning method.
🎭 Subjective Aesthetic Analysis – Assessing perceived natural movement through user evaluation.
🎮 Implementation in Unity – Designed for extensibility and iterative testing.

Implementation & Testing

The methods are implemented within the Unity engine, providing a flexible platform for visualization and testing. The system is built with extensibility in mind, enabling future enhancements and experimentation with alternative movement techniques. Comparative analysis is conducted through controlled simulations and user studies, measuring the perceived quality and realism of each approach.

Conclusion

The research into both methods was a success. As per user study, some of the proposed methods scored better than the state-of-art motion planning algorithm, while also taking incomparably less time to compute.
One of the dynamic methods proposed was ultimately implemented into our simulation at Saab.
The thesis lays a foundation for potential additional research in the future.

Preview

One of the proposed traversal methods.
This one in particular is a dynamic method based on behaviour tree and analytical vehicle model.