Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!wuarchive!uunet!sli!jng
From: jng@sli.com (Mike Gilbert)
Newsgroups: sci.space.shuttle
Subject: Re: Computer Controlled Landing
Message-ID: <1991Mar26.231854.10390@sli.com>
Date: 26 Mar 91 23:18:54 GMT
References: <9103201921.AA17284@cmr.ncsl.nist.gov>
Reply-To: jng@sli.com (Mike Gilbert)
Organization: Software Leverage, Inc., Arlington, MA
Lines: 31

Regarding the computer-controlled landing discussion, and the observation that
the astronauts take over at 100 feet AGL:

I worked at Intermetrics (a NASA subcontractor for the fifth "backup" flight
computer's software) back when the shuttle was being developed.  I had no
personal involvement in the shuttle program, but another engineer who did
explained the landing problem to me as follows:

The AP-101 shuttle flight computer has only single-precision integer
arithmetic.  Using single precision, the Autoland software can't compute its
location with the precision required to do a landing.

For example, it might only know the shuttle's height AGL to, say, +/- 20 feet,
since the least significant bit of the AGL value would represent a 20-foot
difference.  This tolerance isn't a problem for most of the descent, but a
touchdown 20 feet BELOW the surface of a runway would be a big problem.  This
would explain why the astronaut has to take over, but only at the last minute.

Again according to this engineer, the algorithm was re-coded as (software)
double precision, but it then ran too slowly to be real time ("five seconds
after the crash, it would know exactly how it should have landed...").

I cannot personally vouch for this explanation, but the engineer who explained
it to me (back in 1979 or so) should have been in a position to know.

I recall seeing that the new shuttle computers will have hardware double
precision integer arithmetic, so perhaps Autoland will become more useful
when the new computers become operational.

Mike Gilbert
Software Leverage, Inc.