As autonomous mobile robots are increasingly deployed in complex and unstructured environments, acquiring stable and high-quality visual information has become essential for reliable perception. However, mechanical vibrations and rapid attitude variat...
As autonomous mobile robots are increasingly deployed in complex and unstructured environments, acquiring stable and high-quality visual information has become essential for reliable perception. However, mechanical vibrations and rapid attitude variations caused by uneven terrain significantly degrade the quality of vision sensor data. These disturbances adversely affect feature tracking performance, leading to degraded accuracy in vision-based algorithms such as Simultaneous Localization And Mapping (SLAM) and Visual Odometry.
Conventional digital image stabilization techniques rely on post-processing operations such as cropping and warping, which inevitably reduce the field of view and image resolution. In addition, most existing mechanical stabilization approaches based on two-axis gimbal systems are insufficient to compensate for the full three-dimensional motions encountered in mobile robot platforms.
To address these limitations, this thesis proposes a hybrid hardware--software stabilization system based on a three-axis gimbal designed for mobile robotic applications. The proposed system is formulated through kinematic modeling and implemented using a hierarchical control architecture that integrates high-level motion planning with low-level real-time control. A PID controller augmented with a feedforward term is employed to improve dynamic tracking performance and ensure rapid disturbance rejection.
The effectiveness of the proposed system is validated through real-world driving experiments conducted on artificially constructed uneven terrain. Experimental results demonstrate that high-frequency vibrations transmitted from the robot base are significantly attenuated, resulting in an approximate 60\% reduction in angular velocity fluctuations of the vision sensor. Furthermore, image analysis shows that motion blur and texture degradation are effectively suppressed, leading to an approximate 65\% reduction in feature tracking error while maintaining a stable normalized Harris score.
These results confirm that the proposed three-axis gimbal stabilization system enables robust and high-quality visual data acquisition in dynamic environments, thereby enhancing the reliability of vision-based mobile robotic applications.