JAUS rant

I haven't seen it mentioned on here but some of you may know that there
exists a protocol for passing control messages between an operator console
and autonomous vehicle.  JAUS, the Joint Autonomous Unmanned Systems
protocol was designed as a hack by several defense contractors to unify
and standardize the methods used to control autonomous vehicles.

Unfortunately what it was envisioned to be, and what it currently is, are
at opposite ends of the spectrum.  In the past months that I've been
forced to use it I've become more and more disgusted over what I see as
fundamental flaws that could have been avoided if folks who actually
understood computer architecture would have been properly involved.  The
following is my list of gripes and reasons I feel (and hope) that JAUS
dies a horrible death.

1) JAUS is far too abstract a model to map easily to current hardware
components.  While reading the specification documents I had this vision
going thru my head of a couple of rubber room college professors sitting
down and arguing out a method for doing something without any real world
experience related to what they were defining.

2) JAUS is little-endian!!!  Who in their right mind designs a network
transportable protocol using a little-endian number format?  The
implications of this are that programming for JAUS becomes a very large
pain in the ass, especially on machines that use big-endian numbers.  You
must define manipulator/accessor methods for every stinking field in every
type of message allowed for in the spec because you cannot simply map the
data buffer from one machine to another and get meaningful information out
of it.  Again, probably defined by someone who thinks the Intel based PC
is the only type of machine on the planet...and Intel is not an optimal
choice for battery powered robots: too much power draw

3) JAUS has no facility for dynamic discovery of what robotic components
are available in a network or what their attributes might be.  More on
this later, but it makes the idea of a generic interoperable operator
console an utter impossibility.

4) JAUS messages for handling audio-visual or ad-hoc sensor data are
severely lacking or non-existent.  The current hack for video is crippled
because it makes certain assumptions about the format of the video stream
and the capabilities of the camera.

5)  Since starting to use JAUS last August I've known five different
vendors who all had different interpretations of how a single function is
accomplished under JAUS.  This tells me that it should not be used.

6)  The written documentation for JAUS suxs.  I've noted several places
where the specification is vague or outright conflicts with what is
written in a another section.

The reason I bring up this rant in this group is because JAUS is simply a
bad implementation of a good idea.  I realize that many of you are
hobbyists and not professional robotic engineers but I've noted from
reading the posts that you are generally asking the right questions and in
many cases sharing good answers.  I'd like to see folks outside the
defense industry think about this issue of a standard robotics control
language.  I'm convinced the non-military sector can come up with a far
superior scheme.

-noone of consequence