For a specific case where I want to publish odom tfs for a bunch of simulated robots, I am trying to remap the "/tf" topic to robot specific topics such as /robot1_tf, /robot2_tf and so on. More precisely, somewhere along a control loop that simulates all my robots in one common process, I do this in order to feed amcl:
tf::TransformBroadcaster* tfBroadcaster;
void ROSInterface::sendOdomData(const Vector<OdomData> &odomData)
{
for (uint i = 0; i < odomData.size(); i++) // process odomData for every simulated robot
{
// Publish the odom message.
nav_msgs::Odometry odom;
odom.header.stamp = ros::Time::now();
odom.header.frame_id = "odom";
odom.child_frame_id = "base_link";
odom.pose.pose.position.x = odomData[i].x;
odom.pose.pose.position.y = odomData[i].y;
odom.pose.pose.position.z = 0.0;
odom.pose.pose.orientation = tf::createQuaternionMsgFromYaw(odomData[i].theta);
odom.twist.twist.linear.x = odomData[i].vx;
odom.twist.twist.linear.y = odomData[i].vy;
odom.twist.twist.angular.z = odomData[i].omega;
odomPub[i].publish(odom);
// Publish the odom transform with the TransformBroadcaster.
geometry_msgs::TransformStamped odom_tf;
odom_tf.header = odom.header;
odom_tf.child_frame_id = "base_link";
odom_tf.transform.translation.x = odomData[i].x;
odom_tf.transform.translation.y = odomData[i].y;
odom_tf.transform.translation.z = 0.0;
odom_tf.transform.rotation = odom.pose.pose.orientation;
tfBroadcaster->sendTransform(odom_tf);
}
}
See how I have a vector of odomPub for every robot? I can set that up easily like so:
std::vector<ros::Publisher> odomPub;
for (uint i = 0; i < robots.size(); i++)
odomPub.push_back(nh->advertise<nav_msgs::Odometry>(std::string(robots[i].name + "/odom"), 1));
But with the TransformBroadcaster this is pointless and all published transforms are meshed together in one "/tf" topic where it is impossible to distinguish them from each other.
After a lot of googeling, I came across the idea that I could use a separate node handle for every simulated robot and instantiate each node handle with a specific mapping like so:
ros::M_string remappings;
remappings.insert(std::make_pair("/tf", "/robot1_tf"));
remappings.insert(std::make_pair("tf", "/robot1_tf"));
nh1 = new ros::NodeHandle("devices", remappings);
but no. It doesn't do it. The TransformBroadcaster still publishes on "/tf" as if the mappings weren't there. The mapping does work though on topics that I advertise myself. Is there something I am not doing right? Is there a simple way to fix this or remap /tf some other way?
So far the only way I have been able to successfully remap /tf is using remap in a launch file,
<remap from="/tf" to="/$(arg robot_id)_tf"/>
but this would be an absolute nightmare. The number of simulated robots can change any time, ideally even during runtime. Even if I restart the whole simulation, I would then have to dynamically create a launch file that launches a separate node for every robot so that I can remap their /tf individually, not even speaking of getting the odom data from the simulation process into the separate process of every launched node.
Any ideas are welcome.
But with the TransformBroadcaster this is pointless and all published transforms are meshed together in one "/tf" topic where it is impossible to distinguish them from each other.
The overall design of TF is that it represents/manages all established coordinate frames in your entire system. So here you need to distinguish your robots in the TF network by giving reach robot its own frame_id via:
odom.child_frame_id = std::string(robots[i].name + "/base_link"; // match frame_id used in TF
...
odom_tf.child_frame_id = std::string(robots[i].name + "/base_link"; // separate frames in TF
It's similar to how the turtle_tf2_demo spawns multiple turtle bots.
If you were to go the route of trying to isolate the TF messages for each simulated robot to its own topic (like is properly done with Odometry) you'll run into the issue that none of the other ROS tools (e.g. rviz) is set up to process TF data on topics other than /tf and /tf_static. Sure, you could remap the topic names, but then you would never have the TF data for all of the simulated robots at one time (which I'm assuming is what you want).