This repo is the implementation of the following paper:

"Efficient Sampling-based Trajectory Planning with Dual-Layer Probabilistic Intention Prediction for Autonomous Driving in Complex Intersections"

The open-source code will be published here.

To-Do List

• Results [Completed]
• Datasets [Completed]
• Code [In Progress]

• Create a virtual environment using conda (python3.9)

  conda create -n dlii python=3.9

• Activate the virtual environment

  conda activate dlii

• Install python dependency packages via pip. Open the project directory as follows ../setup/requirements.txt, install dependency package.

  pip install -r setup/requirements.txt
  # Speed up (use pip source in China)
  pip install -r setup/requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple

• Clone this repo from the GitHub.

 git clone [email protected]:byChenZhifa/DLII-JSSP.git

• Download the all scenarios data [here] and save it to project directory ./data_scenarios .

  • 8 scenes demo: scenes_demo(8 for planning ablation study).zip
  • all the intersection scenes: scenes_all(800).tar.xz

 cd DLII-JSSP
 mkdir data_scenarios

3. run.

TODO

Note: This framework initiates with a high-layer inference for the target road and subsequently refines this prediction for the target lane.

Fig. 2. Schematic diagram of dual-layer intention inference for road layer and lane layer

Fig. 4. Schematics on the conversion from Cartesian frame to Frenet frame.

Fig. 6. Multiple Lane Centerline Models based on Lane Frenet States inside One Road

Note: Some results of trajectory planning using predefined linear strategies .

If you have any issues with the code, please contact to this email: [email protected]

If you find our work useful for your research, please consider citing the paper.

TODO

• https://github.com/SS47816/fiss_plus_planner