mppi-in-autonomous-driving

Real-time probabilistic inference-based motion planning for autonomous driving with MPPI (Model Predictive Path Integral). For more information 随机轨迹优化方法入门:以MPPI为例

Build Dependencies

• CMake >= 3.20
• C++17 compatible compiler
• CUDA Toolkit (compute capability 7.5+). You can download it from NVIDIA CUDA Toolkit.
• Eigen3 - Linear algebra library
• spdlog - Fast C++ logging library
• yaml-cpp - YAML parser and emitter
• Protobuf - Protocol buffers for data serialization

Tested on WSL2 Ubuntu 22.04🐧

Build and Run

1. Clone the Repository

git clone https://github.com/your-username/mppi-in-autonomous-driving.git
cd mppi-in-autonomous-driving

2. Build

cmake -S . -B build
cmake --build build -j$(nproc)

3. Run

After compilation, the executable file planning_node will be generated in the build folder. Specify a parameter configuration file for it to start the program.

./build/planning_node --config ./config/default.yaml

Visualization with Foxglove

Before viewing the visualization results, you need to manually load the layout file from assets/mppi_layout.json and install the extension from foxglove-gauge-extension.

The system publishes planning and simulation data via Foxglove WebSocket. Open Foxglove Studio and connect to ws://localhost:8765 to online real-time visualize. Additionally, if the save_mcap option is set to true in the configuration file, the simulation data will be saved in the log folder for offline viewing. For information on using Foxglove, please refer to its official documentation Foxglove Docs.

Here are some quick demos🎬.

lane_change.mp4

merge.mp4

turn_left_1.mp4

turn_left_2.mp4

Acknowledgements

MPPI-Generic provides the C++/CUDA library for GPU-accelerated stochastic trajectory optimization, foxglove-sdk provides visualization and data playback, and CommonRoad for simulation interface.