My primary objective, funded by Adept Technology, was to transfer geometric algorithms to Silma's new simulation package that will be used to facilitate Rapid Deployment Automation (see below). I worked closely with John Craig to learn Silma's proprietary language, Sil, and theit new user interface, called RAPID.

I focused on modelling Adept's Flexible Part Feeder. This feeder singulates parts by causing them to drop from one conveyor belt to another. To model how the feeder will perform over repeated cycles with a given part, it is important to predict the statistical distribution of stable poses as that part is randomly dropped.

In 1992, my group developed an efficient algorithm for this problem (Computing a Statistical Distribution of Stable Poses for a Polyhedron, A. Rao, J. Wiegley and K. Goldberg, 30th Annual Allerton Conference on Communications, Control and Computing, University of Illinois, September, 1992). Versions of this algorithm were implemented by my students at USC: Wiegley in 1992, Zheng Yeh in 1993, and Yan Zhuang in 1994.

In fall 1994, I implemented the algorithm in Sil. The input is a CAD model of the part to be fed (we restrict attention to the polygonal convex hull of such parts) and the part's center of mass. The output is all stable poses of the part: For each, the estimated probability that the part will fall onto this pose. Combined with timing estimates for arm motion and camera cycle time, pose statistics allow us to estimate how often the part can be picked up vs. recycled through the feeder. The analysis is quasi-static in that it does not consider the dynamics of collisions or friction.

Once the algorithm was running in Sil, John Craig integrated it into RAPID. Initially, we estimated probabilities for the class of random parts generated by extruding random convex hulls. Next, we extended the algorithm to handle general parts and estimated probabilities for four industrial parts: an orange insulator cap used in previous experiments, a camera shell from Kodak, and two cluster gears from Johnson Controls. Rob Zanutta of Adept ran a batch of insulator caps and Kodak parts through Adept's flexible part feeder to compare actual with predicted statistics. Note that the algorithm assumes that each part drops in isolation while in the feeder parts drop in bulk.

Orange insulator cap (1036 trials)

Pose(Predicted)Measured
1(7 %)2.22 %
2(9 %)5.02 %
3(20 %)19.69 %
4(32 %)27.12 %
5(32 %)45.95 %

The distribution compares reasonably well with the exception of pose 5, which occurs 14% more frequently than predicted. This is the most stable pose in terms of potential energy, so it is understandable that kinetic energy might shift the distribution toward this pose

Black Kodak Camera Shell (627 trials)

Pose(Predicted)Measured
1 (40 %)52.5 %
2 (14 %)4.5 %
3 (7 %)1.9 %
4 (32 %)39.4 %
5 (7 %)1.6 %

As Rob noted, poses 3 and 5 were susceptible to toppling due to inter-part collisions and belt motion.

Other developments:

• I made several trips with Carlisle to pursue a 2-year project with the National Center for Manufacturing Sciences. The aim is to perform further research and development, jointly with industry, in this area. An announcement of this project should be forthcoming in Jan 1995.

• I gave a talk to Berkeley's Robotics and Manufacturing Program and another to Berkeley's Department of Industrial Engineering and Operations Research. I also gave 2 less formal talks at Stanford.

• I regularly attended meetings of John Canny's RISC group at Berkeley and discussed a number of new projects with John, including a proposal for a robot to view museum artifacts which he is now pursuing with Berkeley's Museum Informatics Program. I also worked with Brian Mirtich and Ioannis Emiris at UC Berkeley to apply Mirtich's dynamic simulator and Emris' convex hull algorithm to the problem of part feeder throughtput estimation.

• I gave a talk at the RIA's 2nd Annual Workshop on Flexible Parts Feeding in Cincinatti, where over 80 industrial representatives came to discuss current problems in assembly. Taking the prize for hardest parts to feed: live earthworms, followed closely by false teeth (with subtle nuances between left and right molars).

• I made two trips to General Motors' Technical Center to explore research potential in the area of assembly. I toured 3 assembly plants and suggested a list of 20 potential research topics to Steve Holland's group. We are hoping to begin work on one of these topics this Spring.

• Working with Anil Rao (U. Utrecht) and David Kriegman (Yale), we developed an efficient algorithm for computing Pivot Grasps, summarized in a paper entitled "Complete Algorithms for Feeding Polyhedral Parts using Pivot Grasps" (submitted to the IEEE International Conference on Robotics and Automation; a more extensive version was submitted to the IEEE Transactions on Robotics and Automation). The abstract:

  To rapidly feed industrial parts on an assembly line, Carlisle et. al. (Brian Carlisle, Ken Goldberg, Anil Rao and Jeff Wiegley, ``A Pivoting Gripper for Feeding Industrial Parts'', IEEE International Conference on Robotics and Automation, May, 1994.) proposed a flexible part feeding system that drops parts on a flat conveyor belt, determines position and orientation of each part with a vision system, and then moves them into a desired orientation. A robot arm with 4 degrees of freedom (DoF) is capable of moving parts through 6 DoF when equipped with a passive pivoting axis between the parallel jaws of its gripper. The idea is to grasp a part with 2 hard finger contacts such that it pivots, under gravity, into a desired orientation when lifted and replaced on the table. We refer to these actions as pivot grasps.

  This paper considers the planning problem. Given a polyhedral part shape, coefficient of friction and a pair of stable configurations as input, find pairs of grasp points that will cause the part to pivot from one stable configuration to the other. For some transitions, pivot grasps may not exist. For a part with $n$ faces and $m$ stable configurations, we give an $O(m^2 n\log n)$ algorithm to generate the $m \times m$ matrix of pivot grasps. When the part is star shaped, this reduces to $O(m^2 n)$. We also study a generalization that considers ``capture regions'' around stable configurations. Both algorithms are complete in that they are guaranteed to find pivot grasps when they exist.

• I co-authored 3 other papers this Fall:

  Sensorless Manipulation Using Transverse Vibrations of a Plate, (with K. Bohringer and V. Bhatt of Cornell), Submitted to ICRA in Oct 1994.

  Fixturing 3D Polyhedral Parts with Modular Struts, (with R. Wagner and Y. Zhuang of USC), Submitted to ICRA in Oct 1994.

  Robot Teleoperation via WWW, (with M. Mascha, S. Gentner, N. Rothenberg, C. Sutter and Jeff Wiegley of USC) Presented at the 2nd WWW Conference in October 1994.

• The last paper describes the Mercury Project which makes our industrial robot accessible to anyone with a WWW browser and NII connection. To our knowledge, the Mercury Project is the first system that allows NII users to remotely view and alter the real world via tele-robotics. The WWW "point-and-click" interface allows users to remotely select images by controlling a robot arm, air jet, and camera over the sand-filled terrarium in our lab. Users can excavate artifacts buried within the sand by positioning the arm, delivering a burst of air, and viewing the newly cleared region. Users share their "findings" through an on-line Operator's log. As of January 1994, the system has been accessed by over 25,000 sites from around the world. It was also described in publications such as Newsweek, the LA Times, the Chronical of Higher Education, and the London Times and on a segment of CBS' Up to the Minute (which you may have missed if you weren't watching TV at 3:45 am on Dec 18th...). I co-directed this project with Prof. Michael Mascha at USC and continued to work closely with the other members of the project team via email. In December 1994, we submitted this project for the NII Award in the Education Division.

• I organized a mini-symposium at Adept to explore Open Problems in Industrial Robotics: Theoretical Aspects of Rapid Deployment Automation (TARDA) Held on: Friday 16 Dec 1994, 12-5pm, Adept Technology, San Jose, CA. From the Announcement:

  ``Adept Technology and Silma, Inc are putting a major effort into developing hardware and software to dramatically reduce the time required to configure assembly lines. In a nutshell, the idea is to accurately simulate a small number of paramerized modular components such as robot arms, conveyor belts, flexible part feeders, modular fixtures, light beams, 2d vision systems, etc. which can then be rapidly configured to assemble a product described by a CAD model. We might think of these modules as a "vocabulary" for automation. One challenge is to develop algorithms for simulating and even "compiling" these primitives. In the past year, a number of new problems have emerged. This informal mini-symposium introduced local researchers and students to Adept's concept of Rapid Deployment Automation. Our objective was to identify new theoretical problems with near-term relevance for industrial robotics.''

  35 people attended:

  Bernie Roth, Stanford; Brian Carlisle, Adept; John Craig, Silma; John Canny, Berkeley; Jean-Claude Latombe, Stanford; Mark Cutkosky , Stanford; Moshe Shoham, Stanford; Danny Halperin, Stanford; Lydia Kavraki, Stanford; Michael Goldwasser, Stanford; Hirohisa Hirukawa, Stanford; Tsai-Yen Li, Stanford; James Nielsen, Stanford; Goutam Chatterjee, Stanford; Bill Verplank, Interval; Mike Oitzman, Adept; Felix Buchi, Adept; David Wilson, Adept; Rob Shaw, UCSC; Ioannis Emiris, Berkeley; Eric Paulos, Berkeley; Francesca Barrientos, Berkeley; Brian Mirtich, Berkeley; Dina Berkowitz, Berkeley; Aaron Wallach, Berkeley; David Parsons, Berkeley; Ed Nicolson, Berkeley; Colin Robinson, USC; Yan Zhuang, USC; Jeff Wiegley, USC; Hadi Moradi, USC; Rick Wagner, USC; Ken Goldberg, USC. A summary of open problems is available by request.

• In November, USC nominated me for the NSF Presidential Faculty Fellowship; I wrote several essays and helped to gather the appropriate information for the application via email.

• In December, Stephen Lu, who will be joining USC School of Engineering in January 1995 to Direct the new Center for Automation and Manufacturing Systems, invited me to serve as Associate Director when I return to USC in January.

• In December, I was elected to serve a 3-year term on the Advisory Committee of the IEEE Society for Robotics and Automation.

I would also like to thank Ellis Horowitz of USC for informing me of the possibility for taking a mini-sabbatical, and Howard Moraff from NSF and Dean Len Silverman and the faculty of the USC Computer Science Department for their encouragement. I'd also like to thank my students at USC for not playing too much while the cat was away, and my assistant Delsa Tan for reading my mail and answering a series of panicky requests from afar. And thanks to my family in the area including Adele, Ali, Nina, Jim, David, and Rose; and to my old friends Jonny and Margaret (and now Caitlin), Pammy Aranow, Scott Balcerek, Kristina Robinson, Ovid Jacob, Danny Halperin, Bruce Donald, and the many new friends I made: Max Mendel, Jenny Cool, Tania Miller, Zane Vella, Ann Hess, Chris Vietor, and especially Eric Paulos, who showed me all the great places to drink coffee and talk about machines.

(Ken Goldberg, 2 January, 1995, San Francisco, CA)