Research Project P14:
Self-calibrating robot-vision system

Research Project Goal

Robotic arms generally measure their position with incremental encoders, which measure the rotation of the joint very precisely but do not provide absolute positions. To calibrate the absolute position, there are many different mathods, all of which have significant drawbacks. Since most robots have cameras which can see their own end effectors, calibrating the position of the end effector from the camera data would be very helpful. This calibration would also provide the mapping from the arm's coordinate system to the camera's coordinate system, which is essential for any type of vision-guided manipulation.

Project Scope

Given a set of camera images of an arm, together with the measured joint angles of the arm, estimate the joint angle offsets and extrinsic parameters of the camera with relation to the arm.

This will be done either with markers on the hand, or with the hand holding an easily identified calibration pattern in a rigit grasp.

Two possible extensions are:

• Use the calibration to validate the full kinematics of the arm (e.g. link lengths and axis alignment)
• Use feature extraction to perform this calibration without adding markers to the hand

Sample Data

Robot kinematics (with sample matlab code to generate pose of hand for a given set of joint angles)

Data for a single calibration consists of several pictures of the arm in different positions, together with the measured joint angles at that point. We can gather data with different types of markers to determine what works best.

Tasks

• Calibrate the camera
• Find location of markers on hand in images
• Solve non-linear system of equations to estimate encoder offsets, camera extrinsic parameters, and 3d location of markers on hand.

Research Project Status

student names here

Point of Contact

Eric Berger, berger04@stanford.edu

Midterm Report

not yet submitted

Final Report

not yet submitted