Skip to content

Latest commit

 

History

History
278 lines (230 loc) · 10.3 KB

eval_28032021.md

File metadata and controls

278 lines (230 loc) · 10.3 KB
Raw

Evaluation Run 28.03.2021 Instructions

Installation

  1. Install arena-rosnav according to docs/Installation.md
  • remember the installation in on local_planner_subgoalmode branch, otherwise rosws update may install different packages in src/forks) 
    git checkout local_planner_subgoalmode
  • remember catkin_make always with realse and python3 
    catkin_make -DCMAKE_BUILD_TYPE=Release -DPYTHON_EXECUTABLE=/usr/bin/python3
  • rememeber source in new terminal 
    source devel/setup.zsh
or
source devel/setup.sh
  1. Make sure you are inside your venv when you start the commands
  2. Make sure your python path is correct by sourcing everything before you start the commands
  • echo $PYTHONPATH should output smt like this:
/home/user/arena_ws/devel/lib/python3/dist-packages:
/home/user/arena_ws/src/arena-rosnav:
/home/user/geometry2_ws/devel/lib/python3/dist-packages:
/opt/ros/melodic/lib/python2.7/dist-packages

[EVAL Group1] Run DRL1,2,3 agents:

  1. [Parameter setting] select subgoal mode in the plan_fsm_param.yaml file located at
src/arena-rosnav/arena_bringup/launch/plan_fsm_param.yaml

subgoal_mode:

  • Change to
    • 0: spatial-horizon-wp,
    • 1: time-space-wp,
    • 2: subsampling-wp
  1. [Terminal 1] Start the simulation environment with the specific scenario e.g. map1, 20 obstacles:
roslaunch arena_bringup start_arena_flatland.launch map_file:="map1"  disable_scenario:="false" scenario_file:="eval/obstacle_map1_obs20.json"
  • roslaunch parameters:
    • map_file:
      • map1
      • map_empty
    • disable_scenario: false
    • scenario_file:
      • eval/obstacle_map1_obs5.json
      • eval/obstacle_map1_obs10.json
      • eval/obstacle_map1_obs20.json
      • eval/obstacle_empty_obs5.json
      • eval/obstacle_empty_obs10.json
      • eval/obstacle_empty_obs20.json
  1. [Terminal 2] Wait until all the obstacles in simulation environment are loaded, start the time-space plan manager:
roslaunch arena_bringup timed_space_planner_fsm.launch
  1. [Terminal 3] Afterwards, start the action publisher:
roscd arena_local_planner_drl/scripts/deployment/

python action_publisher.py
  1. [Terminal 4] Start rosbag recording and name it with following convention: planner_map_obstacles_velocity_subgoal_mode
rosbag record -o drl1_map1_obs20_vel01_spatialhorizon /scenario_reset -e "(.*)police(.*)"
  • rosbag parameter
    • name of the bag:
      • [local_planner]_ [map_name]_ obs[number]_ vel[obs_velocity_magnitude]_ [subgoal_mode]
      • example1: drl1_map1_obs20_vel01_spatialhorizon
      • example2: mpc_map_obs10_vel02_timespace
      • example3: teb_map_obs10_vel03_subsampling

Explanation:

  • this command will record all topics necessary for evaluation
  • -e "(.)police(.)": records all topics containing "police"
  • drl1_map1_ob20_vel_01: name of bag file, in this example the drl1 planner was recorded in map1 with 20 obstacles with lin_vel = 0.1
  1. [Terminal 5] Run the run_agent.py script.
roscd arena_local_planner_drl/scripts/deployment/

python run_agent.py --load drl1 --scenario obstacle_map1_obs20
  • change drl planner:
    • --load drl1
    • --load drl2
    • --load drl3
  • scenario_file:
    • --scenario obstacle_map1_obs5.json
    • --scenario obstacle_map1_obs10.json
    • --scenario obstacle_map1_obs20.json
    • --scenario obstacle_empty_obs5.json
    • --scenario obstacle_empty_obs10.json
    • --scenario obstacle_empty_obs20.json

After one test is over (15runs), stop the rosbag command and all other windows and go to the scenario.json files. Change the linear velocity of every dynamic obstacle and/or change your planner/subgoal mode in the scenario.json file and plan_fsm_param.yaml. Afterwards, start from step 1) again. (start simulation and start rosbag command again). If there are any issues while recording and you want to record a new recording, just stop the rosbag command and simulation and start over again. The rosbags will be saved as a new file.

If you want to check the current scenario progress:

rostopic echo /scenario_reset

This will display the reset count. Once the resets reach the max nr of repeats (set up in the json file), the robot should stay at the goal position. Then you can stop the recording.

  1. Once the test runs finished (after 15 runs),
  • stop the rosbag record (rosbag will be saved in the directory where you executed the command)
  • stop the run_agent.py script
  • stop the simulation environment
  • stop action publisher and timespace manager as well
  • change the parameters inside the plan_manager.yaml if neccessary (subgoal mode)
  • change the parameters inside the scenario.json if neccessary (obstacle velocity)
  • start again from step 1 with new scenarios/parameters

[EVAL Group2] RUN TEB,MPC, DWA, CADRL, COL-RL

  1. Install Dependencies: DRL based local planners need some dependencies to be installed in order to run correctly.
    These can be found in the directory: arena-rosnav/arena_navigation/arena_local_planner/env
  • cadrl:
    • pip install tensorflow==1.4.1
    • pip install numpy
    • build ford msg package in your catkin ws: 
      git clone https://bitbucket.org/acl-swarm/ford_msgs/src/master/ 
catkin_make
  • col-rl:
    • pip install mpi4py
  1. [Terminal 1] Start the simulation with the respective local planners by setting the parameter local_planner (ie "teb"):
roslaunch arena_bringup start_arena_flatland.launch local_planner:="teb" map_file:="map1"  disable_scenario:="false" scenario_file:="eval/obstacle_map1_obs20.json"

below commands 2. and 3. not necessary for teb mpc and dwa !

  1. Set which wp-generator to use by setting the parameter in the yaml file . [Parameter setting] select subgoal mode in the plan_fsm_param.yaml file located at
src/arena-rosnav/arena_bringup/launch/plan_fsm_param.yaml

subgoal_mode:

  • Change to
    • 0: spatial-horizon-wp,
    • 1: time-space-wp,
    • 2: subsampling-wp
  1. [Terminal 2] Start the plan manager :
roslaunch arena_bringup timed_space_planner_fsm.launch
  • roslaunch parameters:
    • local_planner:
      • teb
      • mpc
      • dwa
      • cadrl
      • rl_collision_avoidance
    • map_file:
      • map1
      • map_empty
    • disable_scenario: false
    • scenario_file:
      • eval/obstacle_map1_obs5.json
      • eval/obstacle_map1_obs10.json
      • eval/obstacle_map1_obs20.json
      • eval/obstacle_map_obs5.json
      • eval/obstacle_map_obs10.json
      • eval/obstacle_map_obs20.json

For arena local planner, an additional Terminal is needed (first terminal can be set to teb):

roscd arena_ros/scripts/
python arena_node_tb3.py
  1. [Terminal 3] Start rosbag recording and name it with following convention: planner_map_obstacles_velocity_subgoal_mode
rosbag record -o mpc_map1_obs20_vel01_spatialhorizon /scenario_reset -e "(.*)police(.*)"
  • rosbag parameter
    • name of the bag:
      • [local_planner]_ [map_name]_ obs[number]_ vel[obs_velocity_magnitude]_ [subgoal_mode]
      • example1: drl1_map1_obs20_vel01_spatialhorizon
      • example2: mpc_map_obs10_vel02_timespace
      • example3: teb_map_obs10_vel03_subsampling

Explanation:

  • this command will record all topics necessary for evaluation
  • -e "(.)police(.)": records all topics containing "police"
  • cadrl_map1_ob10_vel_01: name of bag file, in this example the cadrl planner was recorded in map1 with 10 obstacles with lin_vel = 0.1

After one test run is over, stop the rosbag command and the simulation and go to the scenario.json files. Change the linear velocity of every dynamic obstacle and/or change your planner/subgoal mode in the scenario.json file and plan_fsm_param.yaml. Afterwards, start from step 1) again. (start simulation and start rosbag command again). If there are any issues while recording and you want to record a new recording, just stop the rosbag command and simulation and start over again. The rosbags will be saved as a new file.

If you want to check the current scenario progress:

rostopic echo /scenario_reset

This will display the reset count. Once the resets reach the max nr of repeats (set up in the json file), the robot should stay at the goal position. Then you can stop the recording.

[Additional Notes]: Verify the simulation

Once the simulation has started, you can check if these three topics are published:

rostopic list
  • /scenario_reset
  • /sensorsim/police/collision
  • /sensorsim/police/odom

When using "teb", "dwa" or "mpc" you need to start the scenario by manually putting a "2D Nav Goal" once. After each reset the goal will automatically set. If everything worked you can continue to step 2)

Record Rosbags

rosbag record -o cadrl_map1_ob10_vel_01 /scenario_reset -e "(.*)police(.*)"

Explanation:

  • this command will record all topics necessary for evaluation
  • -e "(.)police(.)": records all topics containing "police"
  • cadrl_map1_ob10_vel_01: name of bag file, in this example the cadrl planner was recorded in map1 with 10 obstacles with lin_vel = 0.1

After one test run is over, stop the rosbag command and go to the scenario.json files. Change the linear velocity of every dynamic obstacle in the scenario.json file and start from step 1) again. (start simulation and start rosbag command again). If there are any issues while recording and you want to record a new recording, just stop the rosbag command and simulation and start over again. The rosbags will be saved as a new file.

If you want to check the current scenario progress:

rostopic echo /scenario_reset

This will display the reset count. Once the resets reach the max nr of repeats (set up in the json file), the robot should stay at the goal position. Then you can stop the recording.

[Others] Run Ego Planner

In order to test ego local planner:

  1. change the parameter "use_drl" to be false in plan_fsm_param.yaml
src/arena-rosnav/arena_bringup/launch/plan_fsm_param.yaml
  • use_drl false
  1. start simulation environment

You can choose with or wihout scenarios loaded

  • without senarios:
roslaunch arena_bringup start_arena_flatland.launch map_file:="map1"
  • with senarios:
roslaunch arena_bringup start_arena_flatland.launch map_file:="map1"  disable_scenario:="false" scenario_file:="eval/obstacle_map1_obs20.json"
  1. start the time-space plan manager:
roslaunch arena_bringup timed_space_planner_fsm.launch
  1. in rviz set a goal position using FlatlandNav Goal