uos_active_perception

This is a ROS stack for code of the project FLAP for CAOS (Forward-Looking Active Perception for Clutter-Aware Object Search).

A video of the system running on the PR2 robot can be found on the project web page.

More details can be found in the following publication:

Thorsten Gedicke, Martin Günther, and Joachim Hertzberg. FLAP for CAOS: Forward-looking active perception for clutter-aware object search. In Proc. 9th IFAC Symposium on Intelligent Autonomous Vehicles (IAV). IFAC, June 2016.

How to install the system and its dependencies into an empty workspace on ROS Indigo

1. Add the following lines to your .rosinstall:

# The uos_active_perception repo
- git:
    local-name: uos_active_perception
    uri: https://github.com/uos/uos_active_perception.git
    version: indigo
# A fork of the ROS Navigation Stack featuring symmetrical path costs and batch path planning
- git:
    local-name: navigation
    uri: https://github.com/uos/navigation.git
    version: indigo-flap4caos

2. Update the workspace

wstool update

3. Compile all ROS packages

catkin_make -DCMAKE_BUILD_TYPE=Release

---

NOTE: EVERYTHING BELOW THIS LINE IS OUTDATED (ROS FUERTE)

How to run the system with Gazebo

Using the PR2 robot

Execute each of the following steps in a new terminal:

1. Start Gazebo

    roslaunch active_perception_evaluation race_world.launch gui:=true
   

2. Spawn objects of the demo environment and the robot

    roslaunch active_perception_evaluation spawn_race_objects.launch
    roslaunch active_perception_evaluation spawn_pr2.launch
   

3. Start the simulation; tuck PR2's arms; start AMCL robot localization

    rosservice call /gazebo/unpause_physics
    rosrun pr2_tuckarm tuck_arms.py -l t -r t -q
    roslaunch active_perception_evaluation pr2_navigation.launch
   

4. Launch RViz

    rosrun rviz rviz -d $(rospack find race_object_search)/config/object_search.vcg
   

5. Start mapping and view sampling

    roslaunch race_object_search object_search_prerequisites_pr2.launch sim:=true map:=race_extended_inflated fk:=false
   

6. Start object search using default planner parameters

    roslaunch race_object_search object_search_manager.launch robot:=pr2
   

7. Execute object search with 6 target volumes, each with 20% success probability and 5% termination threshold

    rosrun race_object_search object_search_manager_test p table1 0.2 table2 0.2 counter 0.2 shelf1 0.2 shelf2 0.2 shelf3 0.2 min_p_succ 0.05
   

Using the simplified PR2 (floating kinect)

Execute each of the following steps in a new terminal:

1. Start Gazebo

    roslaunch active_perception_evaluation race_world.launch gui:=true
   

2. Spawn objects of the demo environment and the sensor

    roslaunch active_perception_evaluation spawn_race_objects.launch
    roslaunch active_perception_evaluation spawn_floating_kinect.launch
   

3. Start the simulation; start ground truth robot localization

    rosservice call /gazebo/unpause_physics
    roslaunch active_perception_evaluation floating_kinect_navigation.launch
   

4. Launch RViz

    rosrun rviz rviz -d $(rospack find race_object_search)/config/object_search.vcg
   

5. Start mapping and view sampling

    roslaunch race_object_search object_search_prerequisites_pr2.launch sim:=true map:=race_extended_inflated fk:=true
   

6. Start object search using default planner parameters

    roslaunch race_object_search object_search_manager.launch robot:=floating_kinect
   

7. Execute object search with 6 target volumes, each with 20% success probability and 5% termination threshold

    rosrun race_object_search object_search_manager_test p table1 0.2 table2 0.2 counter 0.2 shelf1 0.2 shelf2 0.2 shelf3 0.2 min_p_succ 0.05
   

How to run automatic evaluation

1. cd into empty directory (lots of log output will be placed here)

2. Without any ROS instance running, start the evaluation

    rosrun active_perception_evaluation evman.py "thesis_nomap(N=20, gui=True)"
   

   • Available experiments: thesis_nomap, thesis_withmap
   • N=n adjusts the number of trials
   • gui=False for headless execution

3. Evaluate generated logs

   • cd into experiment subfolder

   • To evaluate and compare initial plans for all planner configurations:

        rosrun active_perception_evaluation eval_plans.py
     

   • To evaluate and compare full search runs for all planner configurations:

        rosrun active_perception_evaluation eval_runs.py ()
     

   • To graphically compare two search runs (e.g., greedy_0 and iw10_0):

        rosrun active_perception_evaluation evaluate_searchrun.py greedy_0 iw10_0