Data Driven Model

Training the model

• Setup the db with the Drone Simuation Setup

  • instructions: https://github.com/TUM-AAS/neural-mpc

• Setup ml-casadi repository

• Setup acados on Linux locally

• set export PYTHONPATH="/home/{username}/neural-mpc/ml-casadi:/home/{username}/neural-mpc/ros_dd_mpc"

• set export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:~/acados/lib && export ACADOS_SOURCE_DIR=~/acados

• python src/model_fitting/train.py --csv data/simplified_sim_dataset/train/dataset_001.csv

• python src/experiments/eval.py

ROS Environment (WIP)

1. Build and start the Docker container:

./run.sh build
./run.sh up

2. Connect Foxglove Studio:

• Open Foxglove Studio
• Click "Open Connection"
• Select "WebSocket"
• Enter URL: ws://localhost:8765
• Click "Open"

3. Add visualization panels in Foxglove:

• Click "+" to add a new panel
• Add "3D" panel to see the car moving
• Add "Plot" panel to see velocities
• Select topics:
  • /car/odom for car position
  • /car/reference_path for the circular path
  • /car/cmd_vel for velocity commands

Stopping the Simulation

./run.sh down

If there is an error on build, prune the docker system.

docker system prune -f