Comparison of Path Planning Algorithms for Autonomous Vehicles Based on UAV-Extracted Roadmaps
MSc Thesis Topic: Comparison of Path Planning Algorithms for Autonomous Vehicles Based on UAV-Extracted Roadmaps
Description:
Navigating autonomous vehicles in wilderness areas such as forests, mountains, and rural terrains presents significant challenges due to irregular road conditions and dynamic environmental factors. Traditional 2D roadmap extraction methods and static path planning algorithms often struggle to handle the complexity of 3D terrains and sudden changes, such as new obstacles or adverse weather. This research aims to compare the effectiveness of 2D and 3D path planning algorithms using roadmaps extracted from UAV imagery and explore reinforcement learning (RL) techniques for real-time adaptive navigation.
Objectives:
2D Roadmap Extraction and Path Planning: Utilize UAV or satellite imagery to extract 2D roadmaps using advanced semantic segmentation models (e.g., D-LinkNet, UNet). Implement and evaluate traditional path planning algorithms such as A*, Dijkstra, and RRT in static environments.
3D Roadmap Extraction and Path Planning: Enhance 2D roadmap extraction with elevation data or digital elevation models (DEM) to construct 3D terrains. Compare 3D path planning algorithms, such as 3D A*, Dijkstra, and RRT-Connect, against their 2D counterparts to understand their performance on complex terrains.
Dynamic Path Planning with Reinforcement Learning: Investigate RL techniques to develop path planning systems capable of adapting to dynamic environments. Incorporate real-time updates for obstacle detection, weather adaptation, and route optimization.
Performance Evaluation: Use metrics like path optimality, computation time, environmental adaptability, and resource efficiency to assess the comparative performance of 2D, 3D, and RL-based path planning methods.
Expected Contributions:
This thesis will provide insights into the trade-offs between 2D and 3D path planning approaches, their feasibility in wilderness settings, and the advantages of RL techniques for dynamic adaptation. The results will contribute to the design of robust navigation systems for autonomous vehicles in unstructured environments.
The applicant will use publicly available datasets and receive technical support from SZTAKI.
Requirements and Technical Skills:
Proficiency in Python or C++ and familiarity with machine learning frameworks (e.g., TensorFlow, PyTorch).
Experience with UAV technology, semantic segmentation, and path planning algorithms.
Knowledge of reinforcement learning and graph-based algorithms.
Ability to work independently and apply theoretical knowledge to practical problems.