F1-tenth ros2-humble simulator

Model Predictive Contour Control Implementation robustness to Computation Latency

• Implemented MPCC in ROS2 simulator MPCC At every iteration, I linearized the dynamics and constraints to formulate the problem as a QP, then used the HPIPM solver to obtain the locally optimal solution. Manual linearization helped reduce the solving time.
• Controller Dynamics Has been modified to consider the computational latency of solver .

Steps to reproduce the results

• clone this reposittory

• Prerequiste : follow this step to install docker and nvidia-docker toolkit docker&nvidia-docker

• from root of directory:

  • image build command : ./scripts/build/f1tenth.sh
  • run the container : ./scripts/build/devel.sh
  • if need to open multiple terminal in already running container do : docker exec -it <name_of_container> bash

• In one terminal run : bash ros2 launch f1tenth_gym_ros gym_bridge_launch.py

• In other terminal run : bash ros2 launch pure_pursuit sim_pure_pursuit_launch.py This should run the pure pursuit with a deafult map

for obstacle(opponent) detection run the following launch file :

• ros2 launch perception perception.launch.py

To Run the Model Predictive Contouring Control

• ros2 launch mpcc_sim mpcc_sim.launch.py

-Currently MPCC subscribe to the "/perception/detection/raw_obstacles" to get the opponent pose but does not do obstacle avoidance

Instructions to Run RRT

To run RRT, follow these steps:

• Publish a drive command: bash ros2 topic pub /opp_drive ackermann_msgs/msg/AckermannDriveStamped "{header: {stamp: {sec: 0, nanosec: 0}, frame_id: ''}, drive: {speed: 0.0, steering_angle: 0.0}}"

• bash ros2 launch f1tenth_gym_ros gym_bridge_launch.py

• bash ros2 launch rrt sim_rrt_launch.py ( other opponent car will be treated as a static obstacle)

Nominal MPCC ( not robust to computation latency : 30 msec )

nominal_mpcc.mp4

Robust MPCC

robusst_mpcc.mp4