Tentabot-DRL: Deep Reinforcement Learning based Robot Navigation in Dynamic Environments using Occupancy Values of Motion Primitives

Tentabot-Heuristic: Reactive Navigation Framework for Mobile Robots by Heuristically Evaluated Pre-sampled Trajectories (Tentacles)

1. Citation

@inproceedings{akmandor2022deep,
  title={Deep reinforcement learning based robot navigation in dynamic environments using occupancy values of motion primitives},
  author={Akmandor, Ne{\c{s}}et {\"U}nver and Li, Hongyu and Lvov, Gary and Dusel, Eric and Padir, Ta{\c{s}}kin},
  booktitle={2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={11687--11694},
  year={2022},
  organization={IEEE}
}

@article{akmandor2021reactive,
  title={Reactive navigation framework for mobile robots by heuristically evaluated pre-sampled trajectories},
  author={Akmandor, Ne{\c{s}}et {\"U}nver and Padir, Ta{\c{s}}kin},
  journal={International Journal of Robotic Computing},
  volume={3},
  number={1},
  pages={47--68},
  year={2021}
}

@inproceedings{akmandor20203d,
  title={A 3d reactive navigation algorithm for mobile robots by using tentacle-based sampling},
  author={Akmandor, Ne{\c{s}}et {\"U}nver and Padir, Ta{\c{s}}kin},
  booktitle={2020 Fourth IEEE International Conference on Robotic Computing (IRC)},
  pages={9--16},
  year={2020},
  organization={IEEE}
}

2. Related Papers

• Akmandor, N. Ü., Li, H., Lvov, G., Dusel, E., & Padir, T. (2022, October). Deep reinforcement learning based robot navigation in dynamic environments using occupancy values of motion primitives. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (pp. 11687-11694). IEEE. [DOI:10.1109/IROS47612.2022.9982133], [arXiv:2208.08034].
• Akmandor, N. Ü. and Padır, T. (2021). Reactive navigation framework for mobile robots by heuristically evaluated pre-sampled trajectories. International Journal of Robotic Computing, 3(1), 47-68. [DOI:10.35708/RC1870-126265], [arXiv:10.35708/RC1870-126265]
• Akmandor, N. Ü., & Padir, T. (2020, November). A 3d reactive navigation algorithm for mobile robots by using tentacle-based sampling. In 2020 Fourth IEEE International Conference on Robotic Computing (IRC) (pp. 9-16). IEEE. [DOI:10.1109/IRC.2020.00009], [arXiv:2001.09199].
• Von Hundelshausen, F., Himmelsbach, M., Hecker, F., Mueller, A., & Wuensche, H. J. (2008). Driving with tentacles: Integral structures for sensing and motion. Journal of Field Robotics, 25(9), 640-673. [DOI:10.1002/rob.20256].

2. Videos

• IRC 2020 - supplementary video
• IRC 2020 - presentation
• IROS 2022 - supplementary video
• IROS 2022 - presentation

3 Installation

The system has been tested with 20.04 (ROS Noetic).

Follow the tutorials to

• install ROS based on the Ubuntu version.
• set up catkin workspace.

To install tentabot, either follow the manual installation instructions, or attempt an automated install with the following instructions (experimental):

3.1 Clone the tentabot repository into the src folder of your catkin workspace, and change into the tentabot directory.

git clone git@github.com:RIVeR-Lab/tentabot.git
cd tentabot

3.2 Run the bash script to automate installing all dependencies.

Note: This script requires superuser and makes system changes. Review the script before running.

chmod +x install_tentabot.sh
./install_tentabot.sh

3.3 Source the workspace, after successfully built.

source devel/setup.bash

4. Simulation

Tentabot features out of the box simulation support for certain robots. You can view the status of this support here:

Tentabot navigation framework is formed by several submodules such as:

a. Simulation Environment (Gazebo)

b. Data Visualization (Rviz)

c. Map Utility Server

• Subscribes to sensor messages (supports PointCloud2 and LaserScan).
• Publishes an occupancy map in octomap_msgs format.

  d. Tentabot Server

  e. Tentabot DRL (Training / Testing)

4.1 Configure the configuration files located in:

• task_nav: World, goal and robot specific parameters. Example: Turtlebot3
• map_utility: User defined local map features and sensor specific parameters. Example: Turtlebot3.
• tentabot_server: User defined Tentabot algorithm parameters. Example: Turtlebot3.

4.1.1 To enable Tentabot DRL, set "drl_service_flag: true" in the Tentabot Server configuration file.

4.1.2 Set the "mode" parameter in the Tentabot Server configuration file as either "training" or "testing".

4.1.3 To continue "training" or start "testing", set the "initial_training_path" parameter in the Tentabot Server configuration file by providing the location of a trained model.

4.2 Open the tentabot_framework.launch file.

4.2.1 Configure the submodule flags.

4.2.2 Set the names of configuration files.

4.3 Run the Tentabot Framework launch file in a terminal window:

roslaunch tentabot tentabot_framework.launch

5. Real Robot

We tested our framework in real robot platforms including Turtlebot2, Stretch and Jackal.

5.1 Run the robot bring-up (Example: Jackal) which includes:

• publishing robot's links and joint state info
• launching sensor drivers.

5.2 Run a localization estimation algorithm (Example: ROS Gmapping).

5.3 Follow all steps in the previous section noting that in step 4.2.1, set the "sim_flag" argument as False in tentabot_framework.launch file.

6. Contributors

Neset Unver Akmandor | akmandor.n@northeastern.edu \ Eric Dusel | dusel.e@northeastern.edu \ Gary Lvov | lvov.g@northeastern.edu | www.garylvov.com \ Hongyu Li | li.hongyu1@northeastern.edu

7. Credentials

Tentabot was developed at the RIVeR Lab, Northeastern University.