ADI201998/PyTorch-Batch-MPC

Repository associated with RAL-ICRA 2022 submission:

"Multi-Modal Model Predictive Control through Batch Non-Holonomic Trajectory Optimization: Application to Highway Driving"

Click on Image to watch it on Youtube

Individual HD video link

Structure

The folder ros_ws/src contains the implementation of approaches: Standard MPC, Batch ACADO over parallel threads, Frenet Frame Planner in C++, and our proposed Multi-modal MPC. It also contains a highway driving simulator and custom ros2 messages used by the packages.

• mpc_car_acado_single: implementation of standard MPC. The problem formulation can be viewed in the code generation file (code_gen.cpp).
• mpc_car_acado: implementation of batch ACADO or multi-threaded ACADO where each thread solves the optimization problem for different goals.
• frenet_cpp: implementation of trajectory sampling based approach: Frenet Frame Planner in C++
• mpc_car_batch: implementation of our proposed multi-modal MPC that is built on Eigen C++ library.
• highway_car: a highway driving simulator where obstacles are motivated by Intelligent Driver Model (IDM).
• msgs_car: custom ROS2 messages that consists of visualization data as well as control input data.
• stats: folder where the simulation data is saved

Dependencies

• eigen_quad_prog

sudo apt install libeigen-quadprog-dev  
sudo ln -s /usr/include/eigen3/Eigen /usr/include/Eigen

• NGSIM dataset

  The dataset is of the I-80 freeway in the San Francisco Bay area. Download the dataset from the link above and place them in ros_ws/src/highway_car/highway/car. The dataset has been taken from the US Department of Transportaion website.

Installation

After installing the dependencies, build our package as follows:

cd your_ws/src
git clone https://github.com/dv367/Batch-Opt-Highway-Driving  
cd your_ws/src/ros_ws/src  
colcon build  
source ./install/setup.bash  

Setting a high-level driving mission

• There are two obstacles settings: obstacles follow Intelligent Driver Model (IDM) or pre-recorded trajectories from NGSIM Dataset
• In each approach folder, you will find config.yaml, set setting to one of the following:
  • Cruise driving in IDM env - cruise_IDM
  • Cruise driving in NGSIM env - cruise_NGSIM
  • Move with high speed and with preference of rightmost lane in IDM env - HSRL_IDM
  • Move with high speed and with preference of rightmost lane in NGSIM env - HSRL_NGSIM

In the first terminal:

• Running our proposed multi-modal MPC

ros2 run mpc_car_batch mpc_node  

• Running multi-threaded-acado

ros2 run mpc_car_acado mpc_node  

• Running standard-mpc-acado

ros2 run mpc_car_acado_single mpc_node_single

• Runinng frenet-frames C++

ros2 run frenet_cpp frenet_car

In the second terminal:

source ./install/setup.bash  
ros2 run highway_car highway_node2    

Miscellaneous

Using the highway_car simulator for your own project:
To be written