ProtoAndroid Progress Report #3 December 8, 1993 David Santos The Robot Group Pa's Paws- Ed Gonzales of IBM has started to work on the Proto-Android's "dextrous" hand system. He is reviewing the project's research files on existing robot hands. I am updating last year's bibliography on the subject from net sources and dragging in the coolest references from the OCLC system. Agreement exists on the following parameters- The hand must serve as a "foot" during quadrupedal locomotion, so it has to be rugged yet light. We are exploring the use of tough foam composites as used in the construction of sport shoes. Tendons would be braided steel and bones of alloy or stiff plastic. General compatibility with the MIT/Utah hand is desired in hopes of using as much existing development as possible. A hand programming language (HPL) consisting of small motion primitives exists. This approach seems usefully extensible to Pa's gross kinematics, amounting to a body programming language (BPL). Several general purpose scripts have been formally outlined for the initial ProtoAndroid demonstrations. While the MIT/Utah hand luxuriates with five fingers, we have simplifed the design in keeping with the creully severe overall engineering strategy for Pa. A three digit configuration reminiscent of a dextrous Saurian has been adopted. It seems that only about five to ten percent of functionality is lost even though forty percent of the complexity is eliminated compared to the MIT/Utah hand. Airhead Lore- A recent open house at Austin Integrated Systems proved to be a great opportunity for the project. On display were the coolest latest pneumatics. I showed around some CAD plans of the 3 DOF joints developed for Pa, which met with appreciation and a spontaneous offer to equip Pa with state of the art pneumatics. This may constitute over ten thousand dollars of hardware we won't have to buy (or steal). The recent jaunt in LA to attend a convention featuring Animatronics and VR netted a bunch of new sources for specialized resources such as servo valves and hyper-real surfaces (see Report from LA, Pulse Dec.2,'93). Shoprats Anonymous Work continues on the body. James Perez and I are fabricating the frame structures that tie the waist, hips, and shoulders together. (We showed off the pieces at a recent presentation at Kealing Jr. High.) A full scale mock-up of Pa's structure has been built out of balsa and foam to assist in the steel layout and fab.. David Hutchings has taken notes for the next round of 3 DOF joint fabrication (wrists, neck) and purchased the billet from which parts will be machined. These will be 1/2 scale versions of the waist, as the shoulders and hips were 3/4 scale. We are considering NC laser cutting now that cross sections are thin enough. Dead Robot Scrolls- Norm Annals continues to draft Pa's schematics and concept illustrations. If you haven't seen this work you're in for a surprise. Not only is Norm's work vital to the development process, but the documents also serve to promote the project to potential donors and satisfy public curiosity. With a complete set of CAD files others will be able to replicate and upgrade the Sorry Machine easily. The Continuing Control Crisis- Two broad currents exist toward a processing architecture for Pa. The first is the ongoing push by Robot Group members to develop microcontroller based systems. This has gone under various guises such as the APE modules by Bill and Alex, Glen's MultiProcessorRobotController (MPRC) concept, and Carlos Puchol and friends' General Purpose Robot Architecture. Additional layers implementing high level control would come later. Microcontrollers are highly self sufficient and cheap little wonders that vastly populate appliances like cars and VCRs. They have become popular among robotics students and enthusiasts. But their smallness imposes limits to functionality and powerful development tools are only slowly coming available. The second approach uses resources evolved for desktop systems doing scientific and industrial data acquisition and process control. Desktop systems provide a vast universe of luxurious and useful capabilities such as GUIs, specialized processing, AI resources, networks, large memories, etc.. Downside- you can spend tens of thousands of dollars on a dream system. The rise of lightweight battery powered notebook computers has made desktop systems practical for mobile robots. The huge investment in development tools and peripheral makes possible the integration of virtually any desired capability from off the shelf products. A theoretically optimal design incorporates the best of both worlds. In fact, convergence is well along. Tony Garcia received an APE board from Alex Iles and Bill Craig to run the two desktop arms for prototyping Pa's pneumatic control. A variety of position encoder, pressure sensing, valving, and motion processing options will be explored with this configuration (see previous report- Pulse ). One task is to create a virtual Futaba servo in pneumatics in order to reuse some APE code at a much higher level of physical power. The development of a superhuman martial arts robot for Rich Walsh of Boulder Colorado is allowing a detailed cross evaluation of control systems. Rich's project requires a non programming user environment with warrantied components and technical support. There are a handful of windows-based icon-driven plug and play industrial data acquisition and process control development systems. The outstanding product on the planet seems to be National Instruments' LabView, a graphical programming environment for PCs, Macs, and Unix workstations. Austin's usual luck is that the company is based here and interested in exploring Robot Group projects using their products. New graphical compilers only approach, but do not quite equal, the code efficiency of traditional language compilers. On the other hand programmer productivity increases in a great leap. Developing a competent robot is so daunting that ease of development may be the key factor toward real progress. Plus, its so cool to see your process running in realtime in an animated dataflow window. Labview makes a wide variety of add-on components that answer most any doubt about realtime speed (DSP boards) or synchronous performance (a real time integration bus jumpered across their add-on boards- RTSI Bus). Only Omega Scientific has a broader product line, but Omega is more of a clearing house than an integral supplier and they're in Stamford CT.. Personae Agent Architecture- Gilbert Andrade and I continue to swat bugs in the multi-media personae interface being built in Foxpro and Visual Basic. Running the creature on a faster system will do wonders. Current thoughts include eventual runtime morphing between FaceSpace primitives, integrating FaceSpace with Iwerks' Vactor, and adopting a standard representation for face states, gestures, and other kinematics. Conclusion (Not)- If progress is possible toward an "impossible" goal then the ProtoAndroid is right on track. If you wish to somehow contribute to this desperate endeavor please do so.
Thanks for flying PolyCosmos. Join us again.