Grand Challenge Pegasus Team: Programming Pegasus Bridge 1 ?

[igorcarron]

Folks,

In a previous post[*] we made an announcement about the release of the
drive-by-wire code for our entry in the DARPA Grand Challenge. We will
do more in the future (including more data and more code). With regards
to our building the full autonomous code, things are going along well.
However, the somewhat large traffic on our web site had us thinking:

Provided we give a good HOW-TO/API, would there be any interest from
anybody to try their code on our vehicle as long as it is in Python and
safe to run ?

Many people think of way to deal with the programming side of
road-following and collision avoidance at 60 mph, but very few have the
time to build a vehicle that can make these concepts a reality. In
order to respond to this, I was thinking of a possibility where
somebody would submit a code, pay $200 and we would try it on a closed
circuit. Then the programmer would be getting all the data attendant to
the vehicle driving itself through the course following their programs?


The pros for us:
- raise some money for the vehicle
- identify potentially better algorithms -> identify people we would
like to associate ourselves with.

The cons for us:
- this is time not dedicated to focusing on the race
- issues with proprietary code/IP
- issues with safety
- coordination with many parties

Right now I am thinking the cons are overwhelming, what do y'all think
?


Igor.

[*]
http://groups-beta.google.com/group...2ecb3e9139d/af28daca5e385af3#af28daca5e385af3

 

[Alex Stapleton]

I'm thinking that with a decent dynamics engine (PyODE?) you could
write a reasonably realistic simulator to test this sort of code on.
Obviously it won't be as good as actually you know, driving a Jeep
around by wire, but it'd be a tad cheaper and more time efficient for
anyone interested in this sort of thing. There the time involved in
actually writing a simulator which you can experiment on your DBW
code on though. You could probably get one going within a year though
I imagine. Unfortunately from what i've seen of it, the Open Dynamics
Engine is less than accurate under some situations. Unfortunately I
believe all the really awesome ones cost huge amounts of money. Just
a thought. I don't have any actual experience with DBW stuff

 

[igorcarron]

Thank you Alex for your feedback.

What I am really proposing is for others to test their software on our
vehicle for the very reason that there are a lot of unknowns that I
cannot believe can directly modeled into an ODE type of engine (as you
call it.) Several reasons for this:

- one piece of data is the images taken from the vehicle, since we do
not use ranging system like lidar, data fusion using imagery and IMU
data is at the center of devising a good algorithm for an autonomous
vehicle. Unless one were to create a world with different illumination,
shadowing (something that would be quite an undertaking on its own), it
would not be a perfect fit to try out a new algorithm. Another data
stream is sound as recorded in the vehicle, I am not sure on how to
generate this type of information.

- an ODE engine would mean that the dynamics of the vehicle and the
terrain would be known in advance. This is rarely the case.

- Our software is being tested on that vehicle already and the vehicle
is ready, it would cost us less to try it out on the vehicle directly.

Thanks for the pointer on PyODE.

Igor.