Windows Live space

 

There were a series of validation test (Validation and Verification) required because of the complexity of the system. The system could not be delivered until it was bought off that is, proved working. First tests were in the laboratory, final tests at Tinker AFB.

 

Lab: Laboratory Verification

 

Facility: Tinker AFB Facility Test

 

We spent more than a year installing and bringing this beast online. The hangar was a hundred yards of the active runaway. For breaks rollerblading on the taxiways with the understanding from the MP's that planes had the right of way!

 

An engineer's first robot is a lot like the first love. Mysterious yet alluring, problematic but desirable. This Niko Gantry was my first. It was built in Mitchell in Indiana by UTC as part of their major endeavor with Ford Motor Assembly. I got it to build a machine vision guided system for palletizing/de-palletizing turbine compressor blades for the Robots 8 show at Cobo Hall in 1984. I had to develop the communication protocol over RS-232 for the Allen Bradley 8200 Numerical Control. Robots are just machine tools in a fancy dress. This particular one talked like a machine tool with M & G codes. The vision system was a Machine Vision International Genesis 2000 (protoype for the Image Flow Computer whose devleopment cost eventually sunk MVI). It was a Multibus I based Unix box that ran an 8Mhz Motorola 68000. We had to put 8 clock cycle wait states for every memory/load and store so the processor would not outrun the DMA controller.

 

Even so it was fast! This was programmed in a mixture of C and C-shell scripting. All the imaging boards in the Genesis were wire-wrapped. The morphological and segmentation algorithms were done in software and were sloooowwwwww.......

 

Steve (plaid shirt on the right worked for AB). Larry on the left was a TV repairman in Bloomongton Ind that did the wiring when he wasn't fixing John Cougar's Big Screen TV.

 

Still it all worked. In 1983 Safeguard funded MVI and Novell. When I left in 1985 MVI was bigger.

 

Every robot that is under servo control requires a kinematic engine in order to compute the control parameters that are passed to the actuators which provide the motion. This particular robot had 9 degrees of freedom. This kinematics engine (that I wrote) used an iterative solver because the orientation degrees of freedom (aka the wrist) did not have their axis of motion intersect at a single point (Pieper 1968).

 

The solver took me a calendar year of development. It was improved from 1 second per path point prototype written in Sil to 100 nanoseconds per path point written in C++ including full redundancy. All Fanuc paint robots use the Hadi Akeel wrist (named after the VP of Engineering at Fanuc who designed and patented it) which has 3 concentric torque tubes driving the ring gear which provided motion. This allowed some symmetry considerations to be exploited to make the solver faster. It was not a completely generic solver so it would not work for many other kinematic found in the paint robot world.

 

I read Dinesh Manocha's PhD thesis (Cal Berkely) looking for a general solver. He used the multipolynomial resultant technique to provide solutions for generalized six degree of freedom robots, and as most industrial robots are 6 degree of freedom devices this might have wide applicability. At the time this technique was not suitable for real time devices because the convergence would not meet the 100 nanosecond requirement of my real time system (30 seconds on the IBM 370). His technique was a special case of the more general technique of solving systems of polynomial equations using homotopy and analytic continuation. Dr. Wampler's Homotopy method is the most rigorous algorithm available. The solution set will also provide all available manifolds for a given data set.

 

Dr. Charles Wampler (GMR, Notre Dame) has published his work in homotopy and analytic continuation for kinematics.

 

I am currently developing the "generic" 6 DOF solver using the GNU Scientific Library and Dr. Wampler's homotopy method. If it is fast enough on cheap hardware (2Ghz $300 PC) I will port it to the Microsoft Robotics Studio. Dr. Wampler's book is an excellent introduction to the kinematics of novel devices like those seen on the cover above.

 

The best book I have read on industrial robotics is the "Bible" by Dr. Richard (Lou) Paul.

 

The LARPS robot was largely built by turning to page 1 chapter 1 and getting started. This book is now out of print but you can find used copies at Amazon. If you work your way through this book you will learn all that is needed to build a robot and control system, you will also fix more than a few typos/math errors but that is half the fun! A more popular version of this is John Craig's book. John, Ric, and Paul (Silma) consulted on the LARPS project assisting with the simulation environment and optical TCP algorithm.

 

 

There are many other books on robot engineering (many imitators) but for an excellent survey these are the best.

 

 

Dr. Paul was issued a patent for an algorithm in path planning in 1997. We had already implemented this algorithm in the LARPS system, sorry dude. We used it for approach/departure control and to find hidden singularities on linear trajectories for the full and redundant 9 DOF system.

 

 

A robot tool center point is a coordinate frame transformation at the end of the serial chain manipulator. This allows every different tool to have a different spatial configuration and for these difference to be reflected in the robot controller path generator and not in user modifications to any path/trajectory. In the case of the LARPS robot the process tool had a linear offset, but the mapping tool had a full 6 degree of freedom homegenous transformation. Since the tooling was optical a technique was developed for calculating the tool center point transformation using the robot and tool plus a constant geometry fixture that was scanned from multiple points and then a regression algorithm was applied to generate the TCP transformation. This was developed in junction with Dr. John Craig at Silma using Cimstation in 1992. Cimstation is no longer sold as far as I know.

 

It is filled with nice images and it documents the technique quite well. Perceptron and Diffracto have used this technique for more than 2 decades to calibrate their sensors for their automotive assembly body-in-white gauges that virtually automobile space frames are now made with. Perceptron now makes a cloud fitting reverse engineering product called Scan Works that makes use of the surface mapping techniques and this tool transformation algorithm.

 

I had a 3 meter standoff 3 meter ambiquity LIDAR sensor manufactured in 1992 for 3D surface mapping of large objects. This system also used a derivative of the tool center point calibration algorithm. The sensor because of it size was fixture mounted in the test workcell. By moving a tool (we actually had the exact same pointer tool made to .001 inch tolerance for calibration) with a robot the sensor was granformed into a metric space that was usable. The LIDAR sensors were very expensive, and relatively fragile due to the rotating polygonal mirrors that generated the 3-D data cloud. The optical "telescope" that generated the phase shift that was calibrated was prone to drift from mechanical vibration and temperature variations. They were never made in high enough volume to be a successful commercial product. Odetics in Anahiem CA also made a family of these scanners that were slight more accurate, better designed, samller and lighter with an active temperature control in the sensor head for minimizing "telescope" problems. They were discontinued in the early '90's for the same reason. A good highly accurate 2-D scanner attached to a ropbot arm gave equally good data at the expense of calibration, motion  and tool changing.

 

Johan Hallenberg did an excellent job in documenting his technique the thesis is very readable, sorry that it is not new. You can download it here. He also implemented the vision algorithms using the Intel OpenCV open source product. It runs very nicely in Ch interpreter from softintegration for those of you that want to script in ANSI C.

 

 

 

 

This robot has eleven degrees of freedom. The U100 scara rides on an autoguided vehicle (AGV) that moves around the aircraft hangar following wire embedded in the floor so it has an x-y degree of freedom with respect to the rest of the robot serial kinematic chain. However the robot is always stowed duing AGV motion so those degrees of freedom can be cancelled out if the AGV stop positions are well known in an exterior coordinate frame of reference like a theodolite.

 

The photo above shows the U100 scara in the right elbow configuration.

 

The U100 scara is capable of linear motion from its' tool centerpoint. The scara positioner has a left elbow and a right elbow solutuion motion for linear cartesian motion. The robot cannot move between solution manifolds in linear motion. The jacobian becomes zero at this point and is in the denominator of the differential position calculation which is thus infinite. As the robot approaches this singular point the velocities of the joints get faster and faster for smaller and smaller incremental positions. At the singularity they would be infinite so obviously that is not a realizable condition physically. To traverse between mainfolds the robot must move in joint motion and stay well away from the singularity.

 

The P160 robot has six solution manifolds. The second rotational axis can be either up or down (it is up in the above picture). The third rotational axis can either forward or backwards (it is forward in the above photo), and the wrist can be either up or down (it is down in the picture above). The diligent reader should try to imagine what the opposite mainfolds would look like and determine if they are physcially useful for robotic operation.

 

 

The Fanuc P-150 robot is pictured in the above simulation in an up-forward-down solution space. The P-160 has the exact same kinematic configuration except the Denavit-Hartenberg parameters are different. It is an excellent exercise in forward kinematics to put this type of a robot in DH format.

 

The "Hadi Akeel" wrist on this family of robots is offset from the centerline of the motion axis. The final three axis do not intersect in a single point thus there is no closed form solution for the inverse kinematics of this type of robot. Rodney Piper provided this in the late 60's for a PhD thesis. Dr. Charles Wampler  has an excellent book on methods of solutions of robots of this type using muli-polynomial resultants. WHen this system was created this was not a feasible solutuion because the computations needs to be run in a few milliseconds in leaner motion. However with today's high speed multi-core CPU's it is probably a tractable solution.

 

Every point in linear motion for the combined scara and P160 device had an infinite number of solutions, so constraints had to be generated to get trealistic solutions. The algorithm developed is very similar to they was in which a human being (also a highly redundant manipulator) approaches points in space without going "singular". The mathematics of the biomechanics of humanoid mainpulators is covered in detail in the excellent book by Ivancevic et al.

 

It is an excellent exercise to figure out what mainfolds are practical for processing as constraints and those that are useless. The answer will depend on what kind of workpiece is being processed robotically. Hint: you might not want to approach a tool centerpoint on the surface of a $100 million dollar airframe from inside the airframe. The controller will have no problem with this, but the owner of the aircraft certainly will!