CALIBER His Stanford PhD system for routing aircraft cable harnesses: one human designer and a team of software agents, 1993–97
CALIBER was the testbed in his dissertation, Computational Support for the Collaborative Design, Routing and Manipulation of Cable Harnesses (Stanford, 1997). It routed harnesses with up to sixty connectors and hundreds of wires, and designers at a large aerospace company tried it. It ran on a Silicon Graphics workstation, and its code is not on the floppies. Two things run on this page: the 2026 browser rebuild of CALIBER, and the small 1992 program on the floppies where the idea starts.
1. CALIBER, rebuilt for the browser in 2026
- 2026 rebuild from the 1997 thesis, not the original code
- 1997 thesis · original ran on an SGI MIPS R4000
- Original source: does not survive
- Same app as
conru.com/thesis/app, copied here
Watch a 35-second recording instead
2. On the floppies: the seed, about 1992
- Original program · his
CONFIG2.EXE, unmodified - By July 1992, the date the disk was copied
- Turbo Pascal
- 933 lines (
CONFIG2.PAS), plus a 524-line first version,CONFIG.PAS
Six cable ends (red circles), four junctions where bundles split, and a cost: the total length of wire. The program nudges one junction at a time until nothing gets shorter. His 1993 conference paper describes the same first step: “analogous to finding the minimum-energy configuration adopted by a web of interconnected rubber bands strung between the end-point locations.”
Watch a recording of this run
What is CALIBER? Cable harness routing in 3-D. The designer drags bundles in a live 3-D view, and automated agents (creators, mutators and combiners) work on copies of the design in a shared pool. In his tests the human and the agents together beat either one alone. Read the thesis (PDF).
What the floppies hold: no CALIBER code. They have its 1993 groundwork: Word for Macintosh files of his ASME paper draft (May 8, 1993), “Design Flow of Cable Harness Design”, “Harness Internals”, a genetic-algorithms proposal, literature reviews and “Thoughts on Routing Methods”. They also have the two CONFIG programs above. CONFIG.EXE doesn’t start on its own: it looks for Borland’s graphics driver in C:\TP\GRAPH. CONFIG2 has the driver built in, so it runs.
About the copy of the rebuild: it is the app at conru.com/thesis/app/, downloaded on October 23, 2026, so the draft works offline. Two small changes let it run from a file on disk. Its script loads as a classic script instead of a module, and its background worker starts from an embedded copy instead of a URL. Nothing else was changed. It starts muted here.
CONFIG2 runs in js-dos 8.5.2 (DOSBox in WebAssembly, GPL-2.0), self-hosted in ../js-dos/. The emulated CPU is slowed (300 cycles) so you can watch the junctions move; on a 1992 PC the search took a second or two. After the first run the screen turns grey: the program clears it with a text-mode call, so the black lines it draws to erase old positions show up. That is the program’s own behaviour, not the emulator’s. Line counts include comments and blank lines.