We have solved for the transformations. This was done by working through the basic transformations and then hand measuring the distances and angles. These hand measured values were used as the initial "guess" by MATLAB's Optimization Toolbox. Using feedback on a large dataset, the toolbox was able to get error down below 2.5cm. This is acceptable because it is well below our thresholds. Still, we will attempt to improve this after the Biclops Head is mounted on a Tripod.
I have taken a data glove apart and stripped out its circuitry. This was done because the sending of finger flexion information was blocked in hardware when the source was out of sight. I have finished building the new circuit and am in the process of finishing the code for the microprocessor (Atmega8) as well as the corresponding driver for linux.
Joshua Southerland & Charles de Granville
We have working code which recognizes pick and place events with different objects. The goal now is to keep track of changes in the environment such as moving goals. This will be easier to do once we have incorporated our robot head's vision system which should be working shortly. We have finished the stereo calibration of the cameras and are now working out the transformations between different coordinate frames.
Joshua Southerland & Charles de Granville
We have finished a program which logs position and orientation data from the polhemus as well as finger flexion and button data from the P5 Data Glove at the same time.
Using Matlab scripts that we recently wrote, it was possible to plot graphs of the error in position from our fingertips to a goal object. We will have orientation potential included in these plots shortly.
Joshua Southerland & Charles de Granville
We are applying machine learning to teleoperation. The current goal is to demonstrate tasks such as picking up objects and then to have the robot learn from our actions. Eventually we plan on accomplishing a system in which a robot is capable of carrying out tasks based off of the operator choosing objects.
One piece of equipment we have been working with is the P5 Data Glove. This glove allows a computer to gather information about the hand's movements as well as finger flexion information. With it we were able to control a robotic arm located at the University of Massachusetts.
We have also developed software for the Polhemus Patriot. This piece of equipment consists of a sensor that provides position and orientation data relative to a source position and orientation. Our plan is to mount this sensor to the P5 Data Glove, which will allow us to determine the position and orientation of the glove while it is in motion.
Software has also been written to control our Biclops robotic head. The Biclops will play an important role in our vision system. We currently have two cameras mounted on the head, which will allow us to determine the position and orientation of objects.
We will be using a combination of supervised and reinforcement learning to accomplish our goals.
Joshua Southerland & Charles de Granville