Enhancing unmanned vehicle navigation safety: real-time visual mapping with CNNs with optimized Bezier trajectory smoothing

Abstract

Purpose

This paper aims to develop a real-time pothole detection system to improve mapping, localization and path planning, reducing vehicle instability and accident risks. Efficient mapping, accurate localization and optimal path planning stand as prerequisites to realizing accident-free navigation. Despite their significance, existing literature often overlooks the real-time detection of potholes, which poses a considerable risk, particularly during nighttime operations. Potholes contribute to vehicle imbalance, trajectory tracking errors, abrupt braking, wheel skidding, jerking and steering overshoot, all of which can lead to accidents.

Design/methodology/approach

Unmanned vehicles constitute a critical domain within robotics research, necessitating reliable autonomous navigation for their optimal functioning. This research paper addresses the gap in current methodologies by leveraging a Convolutional Neural Network (CNN)-based approach to detect potholes, facilitating the generation of an efficient environmental map. Furthermore, a hybrid solution is proposed, integrating an improved Ant Colony Optimization (ACO) algorithm with modified Bezier techniques, complementing the CNN approach for accident-free and time-efficient unmanned vehicle navigation. The conventional Bezier technique is enhanced by incorporating new control points near sharp turns, mitigating rapid trajectory convergence and ensuring collision-free paths.

Findings

The hybrid solution, combining CNN with path smoothing techniques, is rigorously tested in various real-time scenarios. Experimental results demonstrate that the proposed technique achieves a 100% reduction in collisions in favorable conditions, a 4.5% decrease in path length, a 100% reduction in sharp turns and a significant 23.31% reduction in total time lag compared to state-of-the-art techniques such as conventional ACO, ACO+ Bezier and ACO+ midpoint Bezier, Improved ACO, hybrid ACO+ A*.

Originality/value

The proposed technique provides a proficient solution in the field of unmanned vehicles for accident-free time efficient navigation in an unstructured environment.

Citation

Mavi, T., Priya, D., Grih Dhwaj Singh, R., Singh, A., Singh, D., Upadhyay, P., Singh, R. and Katyal, A. (2024), "Enhancing unmanned vehicle navigation safety: real-time visual mapping with CNNs with optimized Bezier trajectory smoothing", Robotic Intelligence and Automation, Vol. ahead-of-print No. ahead-of-print. https://doi.org/10.1108/RIA-03-2024-0076