Stereo Visual Servoing

This work has been carried out by
Enric Cervera (Robotic Intelligence Lab at Jaume-I University),
François Berry and Philippe Martinet (LASMEA at Blaise Pascal University).

Our setup consists of a stereo rig mounted on the end-effector of the manipulator. We consider a simplified configuration where both cameras are parallel, with identical focal length, and the control frame is located in the middle of both camera frames.

The manipulator is a Mitsubishi PA-10 arm.

Attached to the end-effector of the arm is a stereo rig with two miniature CMOS color cameras, linked to a pair of video processors, which deliver extracted visual features at video rate (60 Hz).

The target object consists of a set of five points: four of them are located at vertices of an 11cm square, whereas the fifth point is the center of the square. Points are colored to help to identify each point for tracking.

Examples of servoing tasks

Visual features are stored with the arm in different positions. In order to move the arm from an arbitrary point towards a chosen position, visual features of such position are used as the reference signal in the control loop.

In the first example, the arm moves towards position 0, approximately above the target.

See data plots resulting from the task when three different feature vectors are used:

• crv2p2dm: raw image coordinates
• crv1p3d: 3D coordinates of points (estimated from the stereo images)
• crv1p2dst: image coordinates divided by stereo disparity

In the second example, the arm moves towards position 2, down to the left of the target.

If you have questions or comments about this page,
please contact Enric Cervera at ecervera@icc.uji.es

This page last updated 05/18/2001 - 09:33.