Welcome to the iPlanner: Imperative Path Planning code repository. The iPlanner is trained via an innovative Imperative Learning Approach and exclusively uses front-facing depth images for local path planning.
A video showing the functionalities of iPlanner is available here: Video
Keywords: Navigation, Local Planning, Imperative Learning
This code is released under the MIT License.
Author: Fan Yang
Maintainer: Fan Yang, fanyang1@ethz.ch
The iPlanner package has been tested under ROS Noetic on Ubuntu 20.04. This is research code, and any fitness for a particular purpose is disclaimed.
To run iPlanner, you need to install PyTorch. We recommend using Anaconda for installation. Check the official website for installation instructions for Anaconda and PyTorch accordingly.
Please follow the instructions provided in the INSTALL.md file to set up your environment and install the necessary packages. You can find the INSTALL.md file in the root directory of the project.
Please refer to the Autonomous Exploration Development Environment repository for setting up the Gazebo simulation Environment: Website, switch to the branch noetic_rgbd_camera.
To build the repository and set up the right Python version for running, use the command below:
catkin build iplanner_node -DPYTHON_EXECUTABLE=$(which python)
The Python3 should be the Python version you set up before with Torch and PyPose ready. If using the Anaconda environment, activate the conda env and check
which python
Go to the iplanner folder
cd <your_imperative_planenr_path>/iplanner
Download the pre-trained network weights plannernet.pt here and put it into the models folder. Noted this pre-trained network has not been adapted to real-world data.
You can also collect data yourself either inside the simulation environment or in the real-world. Launch the data_collect_node
roslaunch iplanner_node data_collector.launch
Provide the information for the necessary topics listed in config/data_params.yaml. The collected data will be put into the folder data/CollectedData, and generate folders for different environments that you can specify in config/data_params.yaml under env_name.
For each of the environments, the data contains the structure of:
Environment Data
├── camera
| ├── camera.png
│ └── split.pt
├── camera_extrinsic.txt
├── cloud.ply
├── color_intrinsic.txt
├── depth
| ├── depth.png
│ └── split.pt
├── depth_intrinsic.txt
├── maps
│ ├── cloud
│ │ └── tsdf1_cloud.txt
│ ├── data
│ │ ├── tsdf1
├── data
│ │ └── tsdf1_map.txt
│ └── params
│ └── tsdf1_param.txt
└── odom_ground_truth.txt
You can download the example data we provided using the Google Drive link here.
Navigate to the iplanner folder within your project using the following command:
cd <<YORU WORKSPACE>>/src/iPlanner/iplanner
Run the Python script to generate the training data. The environments for which data should be generated are specified in the file collect_list.txt. You can modify the data generation parameters in the config/data_generation.json file.
python data_generation.py
Once you have the training data ready, use the following command to start the training process. You can specify different training parameters in the config/training_config.json file.
python training_run.py
Launch the simulation environment without the default local planner
roslaunch vehicle_simulator simulation_env.launch
Run the iPlanner ROS node without visualization:
roslaunch iplanner_node iplanner.launch
Or run the iPlanner ROS node with visualization:
roslaunch iplanner_node iplanner_viz.launch
To ensure the planner executes the planned path correctly, you need to run an independent controller or path follower. Follow the steps below to set up the path follower using the provided launch file from the iplanner repository:
Download the default iplanner_path_follower into your workspace. Navigate to your workspace's source directory using the following command:
cd <<YOUR WORKSPACE>>/src
Then clone the repository:
git clone https://github.com/MichaelFYang/iplanner_path_follow.git
Compile the path follower using the following command:
catkin build iplanner_path_follow
Please note that this repository is a fork of the path following component from CMU-Exploration. You are welcome to explore and try different controllers or path followers suitable for your specific robot platform.
To send the waypoint through Rviz, please download the rviz waypoint plugin. Navigate to your workspace's source directory using the following command:
cd <<YOUR WORKSPACE>>/src
Then clone the repository:
git clone https://github.com/MichaelFYang/waypoint_rviz_plugin.git
Compile the waypoint rviz plugin using the following command:
catkin build waypoint_rviz_plugin
Press the LB button on the joystick, when seeing the output on the screen:
Switch to Smart Joystick mode ...
Now the smartjoystick feature is enabled. It takes the joystick command as motion intention and runs the iPlanner in the background for low-level obstacle avoidance.
The params file data_params.yaml is for data collection
- vehicle_sim.yaml The config file contains:
main_freqThe ROS node running frequencyodom_associate_idDepending on different SLAM setup, the odometry base may not be set under robot base frame
The params file vehicle_sim.yaml is for iPlanner ROS node
- vehicle_sim.yaml The config file contains:
main_freqThe ROS node running frequencyimage_flapDepending on the camera setup, it may require to flip the image upside down or notcrop_sizeThe size to crop the incoming camera imagesis_fear_actUsing the predicted collision possibility value to stopjoyGoal_scaleThe max distance of goal sent by joystick in smart joystick model
If you utilize this codebase in your research, we kindly request you to reference our work. You can cite us as follows:
- Yang, F., Wang, C., Cadena, C., & Hutter, M. (2023). iPlanner: Imperative Path Planning. Robotics: Science and Systems Conference (RSS). Daegu, Republic of Korea, July 2023.
This codebase has been developed and maintained by Fan Yang. Should you have any queries or require further assistance, you may reach out to him at fanyang1@ethz.ch