[ros-users] Obstacles Detection with laser data
Nicolás Alvarez Picco
nicolasapicco at hotmail.com
Thu Jan 20 17:00:27 UTC 2011
Thanks for answering.
Yes I am using fake_localization node that gives me the tf between the base_footprint to the map, that is what it does the function ros_link->get_robot_pose();
/////////////////////////////////////////////////////////
pose2d_t ROS_Link::get_robot_pose(){
pose2d_t ret;
tf::StampedTransform transform;
geometry_msgs::TransformStamped msg;
std::string source_frame("/base_footprint");//!!este y el de abajo van!!!
std::string target_frame("/map"); //!! Ver q va bien aca!
try{
ros::Time now = ros::Time::now();
tl->waitForTransform(target_frame, source_frame, now, ros::Duration(1.0));
tl->lookupTransform(target_frame, source_frame, ros::Time(0), transform);
}
catch(...){
ROS_ERROR("Error obtaining robot pose");
return ret;
}
tf::transformStampedTFToMsg(transform, msg);
ret.x = msg.transform.translation.x;
ret.y = msg.transform.translation.y;
ret.yaw = tf::getYaw(msg.transform.rotation);//!! WHY??
return ret;
}
//////////////////////////////////////////////
The robots starts at the origin of the map, so It doesn t look as if the odom drift is important.
I need the integrator for 2 reason:
1. I dont know how using tf or something to get not only the position but also the velocity, and the integrator gives me that information
2. I need a future position and velocity to implement a checker of potentially unsafe states.
Do you know a more efficient way of doing what I am trying to do?
Nicolas
Date: Thu, 20 Jan 2011 08:17:46 -0800
From: bhaskara at willowgarage.com
To: ros-users at code.ros.org
Subject: Re: [ros-users] Obstacles Detection with laser data
If you're computing the robot position using an integrator, it's not going to be in the map frame unless the robot starts off at the origin in the map frame, and even then it would be subject to odometric drift. The fact that you're able to transform the point cloud into the map frame means you have a localization algorithm running, so you can just use tf to get the base pose in the map frame directly.
- Bhaskara
2011/1/20 Nicolás Alvarez Picco <nicolasapicco at hotmail.com>
Sorry you are right!!
I have done that, the problem is that the detection is not quite robust. I am adding the code that I am using, just for u to have an idea.
I am doing all this when I receive a msgs of laser data from gazebo, this is a callback, is it too much code for a callback??
In the function safety_manager->is_state_safe I am just calculating the distance between the new robot s state given by a runge kutta integrator and each of the point of the laser, previously transformed to the global frame ("/map"). Has someone done this in a more efficient way??
Thanks !!!
Nicolas
void Kernel::Callback(const sensor_msgs::LaserScan::ConstPtr& cloudlaser){
try{
tlistener.waitForTransform("/map", cloudlaser->header.frame_id,
cloudlaser->header.stamp, ros::Duration(1.0));
tlistener.lookupTransform("/map", cloudlaser->header.frame_id,
cloudlaser->header.stamp, lasertransform);
pcl::transformAsMatrix(lasertransform, mat);
}
catch (tf::TransformException ex){
ROS_ERROR("%s",ex.what());
}
projector_.transformLaserScanToPointCloud(cloudlaser->header.frame_id,*cloudlaser, cloud,tlistener);
sensor_msgs::convertPointCloudToPointCloud2(cloud, cloud2);
pcl::transformPointCloud(mat, cloud2, pc_toPcl);
pcl::fromROSMsg(pc_toPcl, laser_fchecker);
double dist_states=0,best_avg=0;
if (start_callback) {
pose2d_t robot_pose = ros_link->get_robot_pose();
start_state.pose = robot_pose;
start_state.vel.front = start_state.vel.ang = 0.0;
actual_state=start_state;
current_act.acc.front = 0;
current_act.acc.ang = 0;
start_callback=false;
}
if ((current_act.acc.front==0) && ( current_act.acc.ang==0)){
pose2d_t robot_pose = ros_link->get_robot_pose();
start_state.pose = robot_pose;
start_state.vel.front = start_state.vel.ang = 0.0;
actual_state=start_state;
current_act.acc.front = 0;
current_act.acc.ang = 0;
}
Action control, best_control;
bool aux=false;
metric_t best_result = 0;
for(double fv = -0.8; fv <= 0.8; fv += 0.2){
for(double av =-0.75; av <= 0.75; av += 0.25){
control.acc.front = fv;
control.acc.ang = av;
if ((control.acc.front == current_act.acc.front) && (control.acc.ang == current_act.acc.ang))
continue;
new_state = integrator->integrate(actual_state, control);
if (!safety_manager->is_state_safe(actual_state, new_state, &laser_fchecker)){ // PICS checker and Mapcheker
ROS_INFO("State_No_Safe:\n");
aux=false;
continue;
}
else
aux=true;
best_state = new_state;
best_control.acc.front =control.acc.front ;
best_control.acc.ang=control.acc.ang;
best_control.end_pose=new_state.pose;
}
}
if (!aux){
best_control.acc.front= best_control.acc.ang=0;
best_control.end_pose=actual_state.pose;
best_state=actual_state;
}
vel2d_t cmd_vel;
pose2d_t robot_pose;
ros::Rate loop_rate(250);
current_act= best_control;
robot_pose = ros_link->get_robot_pose();
cmd_vel = calculate_control(current_act, robot_pose);
ros_link->send_cmd_vel(cmd_vel);
actual_state=best_state;
}
where :
bool Safety_Manager::is_state_safe (Phase_State state1, Phase_State state2, pcl::PointCloud<pcl::PointXYZ>* pts_laser){
robot_r = 1;
double dist_obstacles, dist_states,dist_obstacles1;
dist_states = dist2d(state1.pose, state2.pose);
int size_cloud=pts_laser->points.size();
for(int it=0; it<size_cloud; it++){
dist_obstacles=sqrt(pow(pts_laser->points[it].x-state2.pose.x,2)+pow(pts_laser->points[it].y-state2.pose.y,2)); //distancia entre 2 muestras consecutivas
if(dist_obstacles <= 1) {// quiero decir si el centro de mi robot esta a mas de la distancia del r significa q esta en un estado a seguro!!! y 4 metros es el mx alcance de mi laser
return false;
}
}
return true;
}
Date: Tue, 18 Jan 2011 09:33:13 -0800
From: bhaskara at willowgarage.com
To: ros-users at code.ros.org
Subject: Re: [ros-users] Obstacles Detection with laser data
Hi Nicolas, it's hard to diagnose this without more surrounding context and an explanation of what exactly 'not working well' means. But one possibility might be that pts_laser and state2.pose are not in the same coordinate frame. Have you used tf to transform the point cloud into the global frame?
- Bhaskara
2011/1/18 Nicolás Alvarez Picco <nicolasapicco at hotmail.com>
Hi Everybody!!
I want to detect objects, but just their position to avoid them with the laser data. So I am using these lines:
for(int it=0; it<size_cloud; it++){
dist_obstacles=sqrt(pow(pts_laser->points[it].x-state2.pose.x,2)+pow(pts_laser->points[it].y-state2.pose.y,2)); //distancia entre 2 muestras consecutivas
if(dist_obstacles <= robot_r) {// quiero decir si el centro de mi robot esta a mas de la distancia del r significa q esta en un estado a seguro!!! y 4 metros es el mx alcance de mi laser
return false;
}
}
where pts_laser is a pcl_pointcloud and the state2 give me the future position of the robot.
So it is a checker, if there is a point which is closer to the robot than the radius of the robot, so that state is unsafe.
This thing is not working well. So do you have a better solution or an idea??
Thanks
Nicolas
_______________________________________________
ros-users mailing list
ros-users at code.ros.org
https://code.ros.org/mailman/listinfo/ros-users
--
Bhaskara Marthi
Research Scientist
Willow Garage Inc.
650-475-2856
_______________________________________________
ros-users mailing list
ros-users at code.ros.org
https://code.ros.org/mailman/listinfo/ros-users
_______________________________________________
ros-users mailing list
ros-users at code.ros.org
https://code.ros.org/mailman/listinfo/ros-users
--
Bhaskara Marthi
Research Scientist
Willow Garage Inc.
650-475-2856
_______________________________________________
ros-users mailing list
ros-users at code.ros.org
https://code.ros.org/mailman/listinfo/ros-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ros.org/pipermail/ros-users/attachments/20110120/c2703eae/attachment-0003.html>
More information about the ros-users
mailing list