MPN-RRT*: A New Method in 3D Urban Path Planning for UAV Integrating Deep Learning and Sampling Optimization.

Area of Science:

• Robotics and Autonomous Systems
• Artificial Intelligence
• Computational Geometry

Background:

• Unmanned aerial vehicles (UAVs) require efficient path planning in complex 3D urban settings.
• Traditional RRT* algorithms face computational challenges and suboptimal paths in intricate environments.

Purpose of the Study:

• To develop an enhanced path planning framework (MPN-RRT*) for UAVs in 3D urban environments.
• To improve computational efficiency, path optimality, and trajectory smoothness compared to conventional methods.

Main Methods:

• Integrating Motion Planning Networks (MPNet) with RRT* for intelligent sampling.
• Dimensionality reduction by slicing 3D urban terrains into 2D maze representations.
• Applying transfer learning to adapt pre-trained MPNet models to simplified maps.

Main Results:

• MPN-RRT* achieved a 47.8% reduction in planning time and a 19.8% shorter path length in simpler environments.
• Smoother trajectories were observed, with a 91.2% reduction in average acceleration.
• In complex scenarios, MPN-RRT* reduced flight time by 14.2% and path length by 13.9% compared to RRT*.

Conclusions:

• The integration of deep learning with sampling-based planning significantly enhances UAV navigation.
• MPN-RRT* offers a scalable solution for real-time autonomous systems in high-dimensional environments.
• Data-driven approaches can effectively augment classical algorithms for improved performance.