MPPI with Control Barrier Functions (Shield MPPI) for F1Tenth car

Building:

1. Clone this repo
2. This repo uses some custom message types for communication, defined here: https://github.com/vaithak/f1tenth_icra_race_msgs

   git clone [email protected]:vaithak/f1tenth_icra_race_msgs.git
   colcon build --packages-select=f1tenth_icra_race_msgs
   

3. Build the package

   colcon build
   source install/local_setup.bash
   

Running

• Launch the object and tracking subsystems ros2 launch f1tenth_icra_race r3.oo.launch.py
• Launch the spliner and state machine ros2 launch f1tenth_icra_race r3.ss.launch.py
• Launch the shield mppi controller ros2 launch f1tenth_icra_race r3.c.launch.py

Subsystems and workings

• Opponent Detection and tracking:
  • It uses a clustering method with an adaptive breakpoint algorithm to create bounding box
    rectangles for obstacles.
  • Uses a simplified Extended Kalman Filter for Velocity tracking (assumes a single obstacle).
  • Implemented in C++ for realtime performance of 50Hz.
  • Reference: https://github.com/kostaskonkk/datmo
• Spline-based real-time planner and state machine:
  • Real-time updating of local waypoints if an obstacle is detected in front of the car.
  • Decides which side to overtake from and fits a cubic spline to update the waypoints.
  • State machine to decide when to trail vs when to overtake
  • Reference from ForzaETH: https://arxiv.org/abs/2403.11784
• Shield MPPI-based controller:
  • MPPI with control barrier functions for following the raceline and avoiding obstacles.
  • Local repair minimization step using CBFs is only done when the trajectory follows undesired behavior.
  • Reference of Shield MPPI: https://arxiv.org/abs/2302.11719

Working demo on sim

Screencast.from.05-02-2025.10.51.00.PM.webm

Working version on a real car