marcusvinicius178
commented
3 years ago Hi experts. I am using Python API scripts with AutowareAI 1.14 version in integration with the SVL simulator.
It's almost everyhting working correct. With 2 exceptions I had lately:
1 - Regarding Sensors data: It seems that ROS is not having access to the sensor data from SVL simulator, or the SVL simulator perception sensors are not tracking the pedestrian agent:
https://get-help.robotigniteacademy.com/uploads/default/original/2X/c/c183bd36e0fd60b08ec555672ded243f567968cc.jpeg https://youtu.be/o_sXoUI8HG4
In this short video (at the end) you can check that the pedestrian is not recognized in ROS simulator (rviz), just the previous dowloaded map (that is uploaded locally previously through the runtime_interface). Can be maybe missing some python script code to enable sensors communication ( I have set the line: sensors = ego.get_sensors() ) to get sensor data , is this enough? Or is this more probably an error of subscribe to the correct laser topic (correlated to vehicle-sensor.json topics) in rviz?
In order to debug, I tried to add a bigger vehicle:
But as you can check the collision was not avoided. Then I have taken a look into plugins tab in SVL simulator tab, and found the link to get more info about lidar: https://www.svlsimulator.com/docs/simulation-content/sensors-list/#lidar
I checked that the data are published to /point_cloud topic Then I added the point_cloud_1 plugin in rviz and subscribed to this topic. In a hope to see if the map gets updated to spawn in rviz the NPC agent. If you take a look in rviz when the car approachs the NPC, this vehicle is tracked by rviz (its shape is in a dotted pink square). *However I still have the warning in PERCEPTION module, specifically the Occupancy Grid Map : Message: No map received.
Should I subscribe to another topic in occupancy grid map instead the default /semantics/costmap...? Would you have please some hint to fix this detection issue? Or an example of detection already working that you could provide me?
It seems from my perspective that is more related to Autoware ready to use packages with lidar than an SVL issue. Other users also had this problem. And I would solve with something similar to this:
Summarize:On my_sensing.launch add a subscriber to Simulator/Velodyne/Point_Cloud topic, later convert this data to a /LaserScan topic in a way that the nav_msgs/OccupancyGrid topic is able to receive this lidar data and update the map to make ROS/RVIZ recognize the Agents spawned into the simulation.
However I believe that is something already done to use (regarding lidar). Could you please assist me to find out what pkgs/scripts to use? Or code to use in sensing.launch file?
I have found these links that are very closed related to set the occupancy grid map:
Then I tried to create the OccupancyGrid map and had partial success. Besides the car keeps colliding, once the OccupancyGrid map did not recognize the AGENT (car) as an obstacle (It assigned the value 0 = not occupied) to the Agent. I have followed the tutorial of these links step-by-step and launched the launch files which enabled the nodes baive_motion_predict and ray_ground_filter to send data to the topics:
1 - /points_no_ground 2 - /prediction/moving_predictor/objects Also enabled the optional topic: 3 - /tf And feeded this topic with a forced transform from base_link to velodyne (Once I often received errors of this TF in terminal: Lookup would require extrapolation into the past. Requested time 1620083704.499075072 but the earliest data is at time 1620083727.652629207, when looking up transform from frame [velodyne] to frame [base_link] ):
In hope to fix this I have issued:
rosrun tf static_transform_publisher 0.0 0.0 0.0 0.0 0.0 0.0 base_link velodyne 1000
What was curious (debugging with rqt tool and rostopic echo tools) figuring out what topic was not sending data (to track why the OccupancyGrid map was not produced.) I have seen that the node ray_ground_filter (launched by runtime_manager interface) was not enough to enable the topic /points_raw generate the data. It was also necessary to launch the node ring_ground_filter in conjoint with the ray_ground_filter, afterwards the topic /points_raw started to send data.
Other interesting point is that in any moment the topic /prediction/moving_predictor/objects generated data.....maybe for this reason the OccupancyGrid map usually when echoed returned any data, OR when returned just sent data (0,0,0,0,0,0....) and did not recognize the car. ???
Then I still have 2 doubts to close this issue?
1 - How to enable the OccupancyGrid properly (which modules/launch files are required to enable on runtime_manager interface)
2 - Why when the OccupancyGrid was enabled it still did not recognize the Agent
I have recorded videos of the step-by-steps and debugging procedure described above:
A) OccupancyGrid enabled randomly (Not by a logical step-by-step as tutorial said)
B) OccupancyGrid enabled but AGENT was not recognized as busy spot
Does other AD stack presents this obstacle recognition issue too?
Thanks very much in advance
InterKnight
Hi experts. I am using Python API scripts with AutowareAI 1.14 version in integration with the SVL simulator.
It's almost everyhting working correct. With 2 exceptions I had lately:
1 - Regarding Sensors data: It seems that ROS is not having access to the sensor data from SVL simulator, or the SVL simulator perception sensors are not tracking the pedestrian agent:
https://get-help.robotigniteacademy.com/uploads/default/original/2X/c/c183bd36e0fd60b08ec555672ded243f567968cc.jpeg https://youtu.be/o_sXoUI8HG4
In this short video (at the end) you can check that the pedestrian is not recognized in ROS simulator (rviz), just the previous dowloaded map (that is uploaded locally previously through the runtime_interface). Can be maybe missing some python script code to enable sensors communication ( I have set the line: sensors = ego.get_sensors() ) to get sensor data , is this enough? Or is this more probably an error of subscribe to the correct laser topic (correlated to vehicle-sensor.json topics) in rviz?
In order to debug, I tried to add a bigger vehicle:
But as you can check the collision was not avoided. Then I have taken a look into plugins tab in SVL simulator tab, and found the link to get more info about lidar: https://www.svlsimulator.com/docs/simulation-content/sensors-list/#lidar
I checked that the data are published to /point_cloud topic Then I added the point_cloud_1 plugin in rviz and subscribed to this topic. In a hope to see if the map gets updated to spawn in rviz the NPC agent. If you take a look in rviz when the car approachs the NPC, this vehicle is tracked by rviz (its shape is in a dotted pink square). *However I still have the warning in PERCEPTION module, specifically the Occupancy Grid Map : Message: No map received.
Should I subscribe to another topic in occupancy grid map instead the default /semantics/costmap...? Would you have please some hint to fix this detection issue? Or an example of detection already working that you could provide me?
It seems from my perspective that is more related to Autoware ready to use packages with lidar than an SVL issue. Other users also had this problem. And I would solve with something similar to this:
https://answers.ros.org/question/367056/costmap_generator-workbut-it-topic-semanticscostmap_generatoroccupancy_grid-do-not-have-data/?answer=377423#post-id-377423
Summarize:On my_sensing.launch add a subscriber to Simulator/Velodyne/Point_Cloud topic, later convert this data to a /LaserScan topic in a way that the nav_msgs/OccupancyGrid topic is able to receive this lidar data and update the map to make ROS/RVIZ recognize the Agents spawned into the simulation.
However I believe that is something already done to use (regarding lidar). Could you please assist me to find out what pkgs/scripts to use? Or code to use in sensing.launch file?
Thanks very much in advance