[ros-users] [Discourse.ros.org] [Next Generation ROS] ROS2 Navigation - Input requested

[Geoffrey Biggs]

I'm not an expert navigation researcher but I am a user and an experimenter.

For me, the biggest desire is for navstack2 to be a navigation framework built using ROS2 principles rather than a monolithic navigation solution, as it is in ROS1. This means defining the separate components (local planner, global planner, local map probider, global map provider, planner supporters such as costmap providers, rescuers, path follower, and so on), defining the interfaces between them, and specifying those as ROS2 messages, services and actions. We should be able to say "a node (or set of nodes) that provides a global planner compatible with navstack2 should publish/subscribe/use these topics, services and actions and provide these parameters, at a minimum". The key thing here becomes defining the APIs between the different parts of the navigation stack.

Navstack2 should take advantage of the capabilities of ROS 2 to make things nodes but then keep them in the same process so that message passing is nearly cost-free.

Being possible to specify via a configuration file what the global planner is, what the local planner is, etc. would also be possible. This is to have it not just be a bunch of nodes with defined interfaces, but be more of a framework where it is simple to build a complete navigation stack without feeling like you are plugging things together manually.

If we do this, then we can achieve the goal of allowing different planners to be plugged in, etc. that is frequently stated in this thread.

It will also ensure that it is inherently easy to introspect the internal navigation process, because we can intercept all the messages flying around between the different parts.

Building on this, navstack2 should then provide a default configuration that provides an equivalent or better navigation functionality to the ROS1 navigation stack.

As someone who has a strong desire to participate but is something like 1000% over-committed in time, I would also like to see a central place where I can comment on design decisions and implementation choices made when I have time, without losing track of where everything is. Please make a github project so we can make issues and discuss them, if you haven't already.

[quote="Edu, post:11, topic:4884"]
A simple undocking algorithm
[/quote]

I think this is too application-dependent. But, it should be *really* easy make things like this default behaviour for your robot in navstack2.





---
[Visit Topic](https://discourse.ros.org/t/ros2-navigation-input-requested/4884/12) or reply to this email to respond.