/ic-fyp

Final Year Project for BEng Electronic and Information Engineering at Imperial College London

Primary LanguagePython

Distributed Multi-Robot Trajectory Planning in a Warehouse Environment

Final Year Project for BEng Electronic and Information Engineering at Imperial College London. The thesis paper is here.

Folder Structure

  • src
    • multirobot_control: Package containing source files, world files, and launch files. This will be where everything is launched from.
      • multirobot_control: Source files containing the implementation of all things coded in this project.
      • launch: Source files containing what is needed to launch a simulation run.
      • params: Files defining the different scenarios and the parameters for each of them.
      • worlds: Where the simulation worlds are defined. The simulation is built around factory_world2.
    • planner_action_interfaces: Custom ROS2 messages, services and actions defined for the purposes of this project
    • robot_base: URDFs source files for the robot.
  • doc: Pictures required for the report.
  • evaluation: Code to auto test the algorithms inside the simulation.
  • todo.md is a changelog and left to show the progress of this project.

Dependencies

  • Gazebo 11
  • ROS2 Galactic
  • pyyaml

Do install these before attempting to run anything in this project.

Setup

The pathing for this project is hard-coded to my machine. After cloning this repo, I recommend going through it with an editor like VS Code and searching for the string /home/tianyilim/fyp/ic-fyp and replacing it with whatever file path you have cloned this directory in.

Future instructions will assume the base directory of this repo to be $WORKSPACE.

To build the project, cd $WORKSPACE and run colcon build. This builds the project in the ROS build system.

In another terminal, cd $WORKSPACE and run source install/setup.bash, which associates the built files with the ROS launch system.

Usage

Demonstration

To view robots running around, run ros2 launch multirobot_control spawn_bots.launch.py. This will prompt you to choose one of three scenarios:

  • Single robot scenario
    • single-robot
  • Robot swap scenario
    • robot-swap
  • Randomized scenario. The source files for each of them are located in src/multirobot_control/params.
    • scenario_random is where several robots are spawned in random locations with random goals. To choose how many robots to spawn, you can edit this file. Remember to follow the structure of naming robots in order with name as "robotN" with N being an increasing integer. Maximum number of robots supported is 16.
    • Robots and goals will be in these locations startend

By default the scenario will not auto-start. To start it automatically, run ros2 launch multirobot_control spawn_bots.launch.py goal_creation:=true to launch the goal assignment node alongside the rest of the simulation.

To visualize the simulation in the Gazebo GUI, add in headless:=false to the list of arguments.

The list of arguments is as follows:

  • params_file : The path to the parameters file that sets the settings used by the DWA and RRT planners. Defaults to $WORKSPACE/src/multirobot_control/params/planner_params.yaml.
  • headless: true to run without the Gazebo GUI, otherwise false. Defaults to true.
  • log_level: Choose between debug, info, warn, error, based on the desired verbosity. Defaults to info.
  • rviz: true to run with the RViz demonstration, false otherwise. Defaults to true.
  • goal_creation: Whether or not to auto-start the goal_creation so the test scenario automatically starts. Defaults to false.

To change the parameters of the simulation, go into $WORKSPACE/src/multirobot_control/params/planner_params.yaml.

There is also a realtime_factor parameter in each of the scenario files that is currently set to 1.0. If you wish to speed up the simulation, you can increase it. However, do edit gazebo_params.yaml as well. The publish_rate parameter should not be less than 50*realtime_factor.

Evaluation

To test the robots, go into $WORKSPACE/evaluation. There are a few files to be configured:

  • config.py sets the search space based on the parameters. The default is to change the planner type with respect to the number of robots in the simulation. Other scenarios can be tested as well.

  • create_test_scenarios.py creates test scenarios based on the settings in config.py.

    • The settings are set in scenario_settings and param_settings inside the file.
    • scenario_settings sets up the test scenario within the test world.
      • If desired to run random goals, leave scenario_settings['goal_array'] to be an empty list "[]". Else populate it with elements equal to exactly the number of robots in the simulation. An example is given in the comments, but another can be found in $WORKSPACE/src/multirobot_control/params/scenario_swap.yaml.
      • If desired to run random goals, specify the total number of goals and the watchdog timeout as well.
      • Provide a set of starting x, y, and theta values for each robot if desired to start them in defined positions. An example is in the comments.
    • param_settings sets up the default configuration of the robot.
      • This is similar to what is in $WORKSPACE/src/multirobot_control/params/planner_params.yaml.

  • visualise_results.py runs through the results of the tests. However, you will need to modify it to display the relevant metrics you wish. It collates:

    1. Average number of goals completed per robot per run
    2. Total number of goals completed per run
    3. Average distance travelled per robot per run
    4. Average time spent planning per robot per run
    5. Average time spent moving per robot per run

  • evaluate_planners.bash creates test scenarios by invoking create_test_scenarios.py and runs all of them.

  • To automatically run tests, run bash evaluate_planners.bash. Depending on the test, this may take quite a while.

Illustrations:

  • Sim Environment: sim-env
  • Robot Photos side-view top-view
  • Many robots in a simulation many-robots

Results:

  • Total goals per run with increasing robots total-goals
  • Average goals per robot per run with increasing robots total-goals
  • Average time taken per waypoint per robot per run with increasing robots total-goals

Parameters

DWA params

Parameter Name Default Value Description
robot_radius 0.35 The value of $r_r$ , furthest edge of robot from the center.
safety_thresh 0.3 Safety distance $r_s$
simulate_duration 0.4 Seconds to forward simulate a set of control inputs
action_duration 0.03 Seconds to carry out a set of control inputs
linear_speed_limit 0.6 Max value of $v_{lin}$ (m/s)
angular_speed_limit 1.5 Max value of $v_{ang}$ (rad/s)
linear_step 0.1 Step size for $v_{lin}$
angular_step 0.2 Step size for $v_{ang}$
dist_thresh_hi 0.3 Distance threshold from robot to goal to consider to have reached a waypoint
dist_thresh_lo 0.05 Distance threshold from robot to goal to consider to have reached a goal
dist_method "L2" Method of calculating the distance to any point. Currently the only implemented method is the Manhattan distance, or L2 norm.
inter_robot_dist 3.0 Multiples of robot_radius for robots to keep away from other robots.
orientation_ub_deg 180.0 Max orientation difference to be considered in scoring function for heading
orientation_ub_deg 20.0 Min orientation difference to be considered in scoring function for heading
angular_K 1.0 Max proportion of s goal that would be deducted from the DWA score based on the orientation of the robot
goal_K 10.0 Score for distance to goal
obstacle_K 1.0 Cost for distance to obstacles
stall_det_period 1.0 Time window to check for stall
stall_dist_thresh 0.1 Lower distance (m) bound for stall detection
replan_duration 5.0 How long to wait (s) to replan when repeated replan requests are made
move_towards_goal_hint True Whether or not to use the stall detection method to give the robot inital values for command input
Parameter Name Default Value Description
rrt_path_bias 0.1 Bias rate for RRT planner to extend directly towards goal
rrt_it_lim 500 Maximum number of RRT iterations
rrt_it_min 50 Minimum number of RRT* iterations
rrt_max_extend_length 1.5 Distance (m) to extend a new point away from the original point towards a new one
rrt_connect_circle_dist 1.5 Distance (m) to find near nodes in the RRT* algorithm
rrt_debug_plot False Whether or not to plot the pathfinding process to matplotlib
waypoint_skip True Whether or not to skip redundant waypoints
waypoint_replan False Whether or not to replan if waypoints are too far away
local_planner "dwa_replan_server" Choice between DWA-R (default) and base DWA ("dwa_action_server") implementations