Visual Servoing Of Unicycle-like Robots With Eye-In-Hand Monocular Cameras.pdf

This chapter explores visual servoing—guiding robots using only camera feedback—for differentially driven unicycle-like vehicles with limited field-of-view (FOV) constraints. The main focus is on accurate robot positioning using a monocular camera fixed onboard ("Eye-In-Hand" setup).

Key Issues:

Limited FOV: Robots can't always keep observed features within the camera's view as they move.
3D Localization: Extracting useful 3D information from a single, limited-aperture camera is challenging.

Proposed Approaches:

The chapter presents two main visual servoing schemes:

1. Position Based Visual Servoing (PBVS):

Uses an extended Kalman filter for localization, combining robot motion models and odometric data to estimate 3D position.
Aims to directly control the robot's 3D pose based on these estimated parameters.
Advantages: Relatively straightforward implementation.
Disadvantages: Requires accurate odometry and can be less robust when features are occluded.

2. Image Based Visual Servoing (IBVS):

Tracks desired image features directly in the image domain, representing optimal paths for the robot.
Can be further divided into:
Appearance-based: Uses visual maps built from consecutive images without 3D information.
Hybrid: Combines PBVS and IBVS depending on proximity to the target.
Advantages: Potentially more accurate than PBVS when features are visible.
Disadvantages: More complex implementation, requires robust feature extraction algorithms.

Contributed Work:

The authors propose a "Visual SLAM for Servoing" approach combining:

A PBVS control scheme using a hybrid control law to handle limited FOV.
A Simultaneous Localization and Mapping (SLAM) algorithm for enhanced localization accuracy during visual servoing.

This integrated method visually navigates the robot through an extended visual map before reaching its desired goal.