From - Sun Dec 08 11:33:05 2002
Return-Path: <robot@ultimax.com>
Sun, 8 Dec 2002 14:28:36 -0500
Message-Id: <v03007804ba19211e793b@[216.111.21.4]>
In-Reply-To: <3DEF7EA3.7020509@fourmilab.ch>
Date: Sun, 8 Dec 2002 14:30:02 -0500
To: John Walker <kelvin@fourmilab.ch>, "John Nagle" <nagle@animats.com>
From: Robert G Kennedy III <robot@ultimax.com>
Subject: Robert's 1st-cut response to various points

John Nagle wrote:

> Stanford isn't doing anything, Hans Moravec at CMU says
>he's funded by DARPA and can't do anything, and the head of
>the Robotics Institute at CMU is thinking vaguely about it.

Raj Reddy or Takeo Kanade?

Is "Red" Whittaker of Red-Zone Robotics (a CMURI spinoff that cleaned up
Three Mile Island) involved in the contest?

>If they don't enter, we might be able to get cooperation in
>the form of software and experience. Having a non-profit
>front would be helpful in that.

Moravec would certainly be the guy to help us if available. He's pretty busy now. When I was on the Hackers Steering Committee year before last, I invited Hans to do the Fireside Chat at the 2001 Conference. He demurred, explaining that MIPS had just become cheap enough (summer 2001) for him to be able to do a bunch of things he always wanted to do in machine vision. He said he has dropped teaching, speaking, and writing commitments for 2-3 years to go back to the lab and do these things.

Even though Oak Ridge Nat. Lab has downsized their robotics division in a big way, there are still strong contacts between CMU and the Lab and the University of Tennessee. Or used to, at any rate. We can exploit those contacts. It would be worthwhile finding out why powerhouses like CMU or MIT are sitting out this contest. We're not going to get any free help out of ORNL, I can tell you that. They're going to want papers signed and cash deposited up front. Their billing rate would turn your hair white.

Jay Gowdy's dissertation mentions Navlab 5 making 2850 miles cross country under computer steering 98.2% of the time. It would be interesting to learn what were the characteristics of the other 1.8% which necessitated human intervention. Ran into rush hour traffic, for instance? Given that *our* race is a closed course, the two sets of problematic situations may be disjoint. If so, the CMU experience would be useful to us.

> > The optimal strategy might be a Hummer at 15MPH.

VEHICLE ISSUES - In John Nagle's design notes, he indicates that $100K+ for a new Humvee would be a misallocation of resources. I concur. But do note that the Hummer doesn't have to be new, especially considering whatever chassis we select will probably get hacked & chopped severely. Imagine my ancient Mach 1 morphing into Mad Max's ride ... oh, wait a minute, it looks like that already. Anyway, a quick look at Kelly Blue Book reveals

http://www.kbb.com/kb/di.dll/kw.kc.urr?kbb.TN;851038;TN001&37830;suv;p&751;Hummer;1993%20H1&10;HU;B4&

Private Party estimate from Kelly Blue Book:
1998 4-dr H1 wagon w/V8 6.5L diesel, 65K miles, auto, a/c, cruise, power;
GOOD condition: $39K
FAIR condition: $36K

+5 years, 86K miles, 6.2L, no cruise
GOOD condition: $20K
FAIR condition: $18K

Note that the decrement between 5 and 10 years is about -50%, but that the decrement between good and fair is only -10%. Recommend older vehicle in good condition. Options like cruise control can be retrofitted from aftermarket sources, but see hybrid powertrain note next.

Hybridizing the powertrain, i.e. from diesel to diesel-electric with individual motors on each wheel (like the Prius) should be considered for the following reasons:

(a) constant speed I.C. engine is more reliable and efficient, eliminates need to directly control I.C. process

(b) more efficient fuel consumption means less initial mass or longer range with same fuel, permits skipping the unnecessary complication of autonomous refueling (see para below)

(c) energy density advantage of liquid fuel is preserved

(d) small I.C. engine, no tranny, means less mass in the powertrain

(e) directly controllable parameters - electric drive - eliminate at least one layer of transducers and associated problems

I understand that this contest is an AI problem, not an automotive design problem, and that hybridizing the powertrain is an major modification which could eat up all the cost savings.

It is better to skip the autonomous refueling stop(s) at checkpoint(s). In 1993-1995, I developed robot-assisted resupply/refueling concepts (POMCUS ammo modules/tools/fuel drums) for the Army. I learned that a full organic (onboard) supply was safer, and that some essential manual dexterity could not be replicated robotically without major penalties in mass and complexity.

A fully enclosed conditioned space is better than trying to ruggedize, heatproof and dustproof the CPU. (However, Lee Felsenstein is building fanless hand-cranked or solar powered wireless computers for deployment in Laotian villages - he might have something to contribute. Like Ed Hass, Lee actually builds things.) Korolev took the approach of full-enclosure with Yantar by putting cameras and controls *inside* with an atmosphere, and

Russkii spysats still follow it. The GMC pickup John mentions could achieve the same thing with a camper shell. Additional fuel canisters if required should be stored externally, so the pickup style of Humvee may be better than wagon.

> > Do you know if dry sump diesel engines will run upside-down for
> > an extended period? Think a chassis with monster truck tires
> > (we spit on silly 2 metre obstacles) projecting above and below.
> > Flip us over--we just keep on rolling as long as the oil flows.

> Bruce Baumgard liked the flippable vehicle too. The
>PackBot ("http://www.irobot.com") works like that, but it's
>smaller. One option along those lines would be one of those
>little offroad buggies equipped with a really big roll cage,
>so that if it rolls over, it ends up upright. But
>realistically, if we have to go 350 miles, we probably
>have to go with something that routinely offroads for
>350 miles.

Battlebot folks like Ed Haas <steamer@sonic.net> whom I'd like to talk to could advise how to make a large vehicle Humvee withstand rollovers. But choosing a Hummer may obviate the rollover requirement, because it's already got a ridiculously low center of gravity, and it's designed to handle 60% grades, 1-meter ditches and 60-cm obstacles, as I recall from the initial RFP many many moons ago.

> Bruce suggests "turtle" and "rabbit" scenarios. The

I like the dual mode idea. The race will be won/lost on who makes up the most time on the high-speed road portions.

> I'm currently thinking in terms of a 2003 GMC truck with
>the StabilTrak active stabilization system using rate gyros,
>accelerometers, and differential braking. If we can tap into
>that system (and I just got the name of somebody who does
>stuff like that) we can get it to tell us how happy the
>chassis is with the current level of driving aggressiveness.
>We can then move between turtle and rabbit mode as indicated.

Notwithstanding all I wrote about Hummers above, there's a lot to be said for just buying an off-the-shelf platform. I'll need to find out if the GMC hardware bus and protocols are compatible with the Common Robotic Platform.

IMAGERY ISSUES - Once the rough race course is bounded (in February, correct?), we could plan a bunch of detailed sub-routes far in advance, based on 1-meter imagery and other sources, as well as defining blocs of terrain to be avoided. Then when the exact waypoints are announced, we quickly patch the appropriate pieces together. However, the raw imagery costs to support this planning could be greater than the vehicle - even that old Soviet 2-meter imagery I got for the Commercial Remote Sensing hearings back in 1994 ran about $10/km^2. A 350 km x 50 km bloc @ 1 meter GSR could run in excess of $100K. A colleague at ORNL is getting recent pricing schedules with volume discounts for me; I'll have it tomorrow. The highest commercially available resolution is now 61 centimeters, from the Digital Globe's QuickBird. 61 centimeters happens to be conveniently close to the size of obstacle the Hummer's specced to cope with. However, the best thing about 61-cm imagery is that it may drop the bottom out of 1-meter prices.

ROUTING ISSUES - They race will almost certainly begin in the high desert (Mojave) outside Los Angeles past where I-15 to Las Vegas climbs the Cajon pass. Say the Victorville area near the decommissioned George AFB, or possibly Fort Irwin, the National Training Center. (I used to commute there 2-3 times a week from Douglas Aircraft in Torrance - arrrgh!) There's no roads closer in to Los Angeles which can (a) be closed to the public, and (b) are navigable in a pickup. A southern route via Twentynine Palms in the low desert is also possible but crosses much more civilization.

PERSONNEL ISSUES - Who will be in charge of recruiting personnel for the team? Do I have permission to reach out to Ed? As you know, he actually builds powered vehicles and engines. He has a nice machine shop up there in Santa Rosa with ample land to accommodate test driving, assuming our entry in this contest will be headquarted in the Bay Area.

How about Lee Felsenstein?

I also have on tap Michael R. Johnson, PE, a mechanical engineering classmate of mine from Cal Poly, now at JPL, who has designed actual flight hardware which has flown to Mars, Jupiter, and Saturn. He also builds animated Rose Floats for the Rose Parade. He helped me found Ultimax. And John Draper, Ph.D., the best human factors guy at the National Lab here; now independent after the downsizing. Knows his way around military SBIRs.

Robert Kennedy, PE

http://www.ultimax.com