Add landmark-based localization examples
Closed this issue · 5 comments
Feature description
As of #268, Beluga features a few sensor models that use landmarks for localization. Alas, we have no example nor demo making use of them -- only unit tests. It would be nice to add a one for each (there are two), even if on a much simplified robot model (aka a box with wheels).
Implementation considerations
This seems like a good opportunity to turn beluga_example
into beluga_examples
, though our 2D simulation in flatland
must give way to a 3D simulation using Gazebo (or Webots or any rigid body simulator supporting camera sensor models).
There are demos now in a branch of the kobuki repo. It's only there because I started from the infrastructure for that repo, but it's actually a different robot, different settings and mostly different worlds.
We need to polish them and create a new public repor for them (we need that for RSS anyway).
https://github.com/ekumenlabs/kobuki_ros2_stack/tree/glpuga/beluga_demo
@glpuga beat me to it. We can use those exact same demos as a basis, replacing the simulation with a simplified, public one. How do you all feel about using Andino's to avoid building and maintaining yet another simulation? Or alternatively, how do you all feel about taking ROS out of the equation and just relying on an in-code simulation using MuJoCo? That one's perhaps more complex but it should make for simpler Beluga examples (Open AI Gym style).
FYI @larodriguez22. This is what we were talking about last time. Sorry for the delay. I'll assign you once I'm allowed to.
@glpuga @larodriguez22 considering the soon to be public beluga-demos
repository already includes landmark-based localization examples, I'm inclined to reassign this ticket, so we can call it a day when beluga-demos
is out. We can open more tickets targetting specific demos (e.g. one with Andino, one that doesn't use ROS, etc.). In fact, perhaps beluga-demos
is the right place for such tickets.
https://github.com/Ekumen-OS/beluga-demos made it out some time ago, so we can call this done.