ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
0

tf2 patterns for high update rates

asked 2019-04-03 06:31:38 -0500

Geoff gravatar image

I'm looking for a good pattern for using tf2 to provide two things:

  1. A set of static transforms for the robot that define things like the wheel positions and sensor locations. These are fixed but they need to be used in a variety of nodes. The transforms come from URDF mainly but there may be some calculated from parameter input when a node starts or from a launch file.
  2. A high-update-rate transform from the world frame to the robot's frame, produced and used only in a small number of nodes.

The problem we have is that we want to be able to get the static transforms in a lot of nodes, but we don't want the overhead of distributing the high rate transform to every node, just the small number that need it. I hope someone has a nice solution within tf2 for this, but I'm happy to hear any other suggestions.

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
0

answered 2019-04-03 06:37:13 -0500

gvdhoorn gravatar image

updated 2019-04-03 06:48:34 -0500

A set of static transforms for the robot that define things like the wheel positions and sensor locations. These are fixed but they need to be used in a variety of nodes. The transforms come from URDF mainly but there may be some calculated from parameter input when a node starts or from a launch file.

You're probably aware of it, but "static transforms" in TF2 are broadcast using a latched publisher, which means they won't be periodically published at all, but rather retrieved once and then stored in clients (ie: buffers). See wiki/tf2: Adding static transform support

All fixed joints in URDFs/XACROs are broadcast as static transforms. static_transform_publisher from tf2_ros also only publishes static transforms.

Under the hood, two topics are used: /tf for regular TF broadcasts (things that change dynamically) and /tf_static (transforms that don't).

The problem we have is that we want to be able to get the static transforms in a lot of nodes, but we don't want the overhead of distributing the high rate transform to every node, just the small number that need it.

There are various approaches one could think of, but remapping might be an easy one: instead of broadcasting the transforms with high update rate onto /tf, remap it to some other topic. Then make the "small number that need it" subscribe to that topic, instead of the regular /tf (remapping again). That should prevent uninterested nodes from being spammed by the high-rate broadcasts, while leaving everything else unchanged.

Some tricks with muxing might be needed to make sure that those high-rate nodes would also still get the regular broadcasts on /tf.

edit flag offensive delete link more

Comments

It would perhaps be nice if tf2_ros::TransformListener could subscribe to multiple, user defined topics carrying TFMessages, as it would avoid the need for muxing in a remapped setup. But that is not supported right now.

gvdhoorn gravatar image gvdhoorn  ( 2019-04-03 10:21:52 -0500 )edit

Remapping sounds like the way to go. It's not ideal but it will do the job of isolating the high-rate data from everything else.

Geoff gravatar image Geoff  ( 2019-04-03 18:06:22 -0500 )edit

Another approach could be to subsample /tf for nodes that don't need high update rates, such as was done in the DRC. Would probably still need remapping (to prevent msgs from being broadcast on /tf at a high rate still).

Such subsampling was done in combination with either custom UDP bridges or with multi-master setups where the nodegraph locally ran high rate versions of topics, but only forwarded msgs to another nodegraph at a lower rate.

Doable, but complex, especially if you start to create many "islands" in this way.

gvdhoorn gravatar image gvdhoorn  ( 2019-04-04 01:35:55 -0500 )edit

Question Tools

3 followers

Stats

Asked: 2019-04-03 06:31:38 -0500

Seen: 397 times

Last updated: Apr 03 '19