Enhanced Visual SLAM for Collision-Free Driving with Lightweight Autonomous Cars.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Autonomous vehicles require robust obstacle avoidance systems.
• Efficient onboard processing is crucial for real-time decision-making in self-driving cars.
• Integrating perception and planning is key for safe and dynamic navigation.

Purpose of the Study:

• To develop a vision-based obstacle avoidance strategy for lightweight self-driving cars.
• To enable the system to run on a CPU-only device using a single RGB-D camera.
• To achieve safe, stable, and efficient path planning.

Main Methods:

• Visual perception using ORBSLAM3 enhanced with optical flow for pose estimation and scene texture analysis.
• Path planning combining control Lyapunov function and control barrier function in a quadratic program (CLF-CBF-QP).
• Obstacle Shape Reconstruction Process (SRP) integrated with CLF-CBF-QP for trajectory generation.

Main Results:

• The proposed method effectively avoids obstacles in complex indoor environments simulated in Gazebo.
• The algorithm demonstrates superior performance compared to benchmark methods.
• Achieved more stable and shorter trajectories in various simulated scenarios.

Conclusions:

• The developed vision-based system provides an effective solution for obstacle avoidance in lightweight self-driving cars.
• The CPU-only implementation makes the system accessible for resource-constrained autonomous platforms.
• The combination of advanced perception and planning techniques ensures robust and efficient navigation.