Thanks for everyone's feedback on this. I was not aware that a version
of the local planner for ackermann navigation existed. Austin, if you
think your solution is reusable, would you mind adding your solution
as one of the answers here:
http://answers.ros.org/question/27/how-can-i-use-the-navigation-stack-on-a-carlike
(I suspect this would directly follow up on Chad's answer)

The initial port of our navigation code will be completely independent
of the navigation stack. However if we can decide on a common
interface, then with a bit of effort we may be able to choose between
different navigation solutions. Stefan, I for one would definitely be
interested if you moved your solution to open source. We probably
won't be working with a terrain as difficult as RoboCup Rescue, but it
might be useful nonetheless to take a look at it.

Jack, what is the best place on the wiki to setup a group for this?

Thanks
Piyush

On Sat, Jan 28, 2012 at 11:18 AM, Jack O'Quin <jack.oquin@gmail.com> wrote:
> On Sat, Jan 28, 2012 at 10:52 AM, Stefan Kohlbrecher
> <stefan.kohlbrecher@googlemail.com> wrote:
>> Hi,
>>
>> we use a Ackermann-driven based UGV for search and rescue scenarios.
>> It is based on a Kyosho Twinforce RC chassis (image for example here:
>> http://www.gkmm.tu-darmstadt.de/rescue/?q=node/8 ). We also use the
>> sbpl_lattice_planner for generating motion plans, but found the
>> navigation stack to be suboptimal for our application, as the vehicle
>> dynamics plus harsh terrain of the NIST standard arenas used for
>> RoboCup Rescue violates some of the assumptions made in the navigation
>> stack. For a 100kg+ platform moving on flat ground, the assumption
>> that odometry provides a stable short term reference frame for example
>> is very reasonable. For a <10kg vehicle going over pitch/roll ramps
>> OTOH there basically is no trustable odometry (see for example 1:05 in
>> this video: http://www.youtube.com/watch?v=nI1DWboC73w ,shows only
>> reactive driving as our planner back then did not work correctly)
>> So our current approach is to use SBPL based plans and not use a local
>> planner as provided by the navigation stack. The whole plan (generated
>> from SBPL or another planner) is sent to the vehicle controller, which
>> then closely follows it. The low level control system of the UGV
>> incorporates  a EKF-based INS that fuses data from inertial/magnetic
>> (IMU) sensors, SLAM pose estimates and wheel encoders and thus can
>> follow the plan nicely. I'll try to make a video of it in simulation
>> in the next few days. The code is not available in our open source
>> repository, but that might change if interest is there.
>
> I think there is definitely enough interest to justify creating an
> Ackermann steering interest group. There are many aspects to it,
> including navigation, interfaces and simulation.
>
> As a start, I suggest creating a wiki page to share links to code and
> techniques. Perhaps it could evolve into a project team to develop and
> maintain ROS packages for dealing with that type of vehicle.
> --
>  joq
> _______________________________________________
> ros-users mailing list
> ros-users@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users