Real-time path planning for autonomous vehicle off-road driving.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Autonomous driving is a rapidly advancing field with significant societal and economic benefits.
• Fully autonomous off-road navigation remains a challenge due to complex terrain and perception requirements.
• Existing autonomous vehicle (AV) systems require sophisticated data processing for sensing, mapping, localization, and path planning.

Purpose of the Study:

• To propose an original approach for autonomous off-road driving by integrating semantic segmentation and real-time route planning.
• To design a modular framework for assistive driving in off-road scenarios, optimized for resource-constrained devices.
• To enhance scene understanding and path prediction capabilities for off-road autonomous navigation.

Main Methods:

• A deep neural network performs semantic segmentation on Red-Green-Blue (RGB) camera images.
• A traversability module fuses Light Detection And Ranging (LiDAR) data with segmentation results to create an occupancy grid map.
• The Rapidly-exploring Random Tree (RRT) algorithm is adapted for real-time path planning.

Main Results:

• The proposed semantic segmentation approach achieved 85.9% precision on the Freiburg Forest Dataset (FFD) and 79.5% on the RELLIS-3D dataset.
• The system demonstrates faster computational times for path planning compared to existing methods.
• The modular framework effectively provides scene understanding for autonomous navigation.

Conclusions:

• The developed approach offers a promising solution for autonomous off-road navigation.
• The integration of semantic segmentation and RRT-based path planning enhances AV performance in challenging environments.
• The system's efficiency makes it suitable for resource-constrained off-road autonomous driving applications.