TERP: Terrain Elevation-based Robot Path Planning

We present a novel formulation to reliably navigate a ground robot in uneven outdoor environments. Our hybrid architecture combines intermediate output from a DRL network with attention with input elevation data to obtain a cost-map to perform local navigation tasks. We generate locally least-cost waypoints on the obtained cost-map and integrate our approach with an existing DRL method that computes dynamically feasible robot velocities.

A video summary and demonstrations of the system can be found here

Dependencies

This implementation builds on the Robotic Operating System (ROS-Melodic).

• Grid Map(grid map library for mobile robots)
• Octomap
• ARL Unity Simulator

Environment

1. Create a Conda Environment

conda create -n terp python=3.7
conda activate terp
conda install pytorch cudatoolkit -c pytorch

2. Modify Grid Map launch file

Goto grid_map/grid_map_demos/launch/ and modify the octomap_to_gridmap_demo.launch file as follows,

  <node pkg="octomap_server" type="octomap_server_node" name="octomap_server">
  <param name="resolution" value="0.5" />

  <param name="frame_id" type="string" value="husky/base" />
  <param name="base_frame_id" type="string" value="husky/base"/>
  <!-- maximum range to integrate (speedup!) -->
  <param name="sensor_model/max_range" value="10" />
  <param name="latch" value="false"/>
  <!-- data source to integrate (PointCloud2) -->
  <remap from="cloud_in" to="/husky/lidar_points" />

  </node>

3. Installing TERP

To build from source, clone the latest version from this repository into your catkin workspace and compile the package using,

cd catkin_ws/src
git clone https://github.com/kasunweerkoon/terp.git
cd ../
catkin_make

Get Started

1. Launch ARL Unity Simulator with Husky

roslaunch arl_unity_ros_ground simulator_with_husky.launch

2. Launch Octomap to Grid Elevation Map Converter

roslaunch grid_map_demos octomap_to_gridmap_demo.launch

3. Run the DWA Planner

rosrun terp dwa_pozyx_goals.py

4. For Testing

conda activate terp
rosrun terp local_waypoint_planner.py

5. For Training

conda activate terp
rosrun terp main_ddpg.py

TERP Paper

Thank you for citing our TERP paper if you use any of this code:

@INPROCEEDINGS{9812238,
  author={Weerakoon, Kasun and Sathyamoorthy, Adarsh Jagan and Patel, Utsav and Manocha, Dinesh},
  booktitle={2022 International Conference on Robotics and Automation (ICRA)}, 
  title={TERP: Reliable Planning in Uneven Outdoor Environments using Deep Reinforcement Learning}, 
  year={2022},
  volume={},
  number={},
  pages={9447-9453},
  doi={10.1109/ICRA46639.2022.9812238}}