Residual Neural Network for Dubins Dynamics Vehicle


Jan 28, 2025 ・ 0 min read

Developed at TU Delft, this project focused on training a Residual Neural Network (ResNet) using PyTorch to learn the dynamics of the Dubins car, enabling efficient trajectory prediction.



Key Features:

  • ➡️ Dataset Generation:   Simulated 10,000 Dubins car trajectories, storing state transition pairs for training.
  • ➡️ ResNet Architecture:   Implemented a residual neural network in PyTorch to model the car's nonlinear dynamics.
  • ➡️ Training Process:   Optimized the network using supervised learning to minimize trajectory prediction errors.
  • ➡️ Trajectory Prediction:   The trained model efficiently forecasts the car’s future states based on initial conditions.

    • By combining machine learning with kinematic modeling, this project demonstrates how ResNet can accurately predict the behavior of a Dubins car for trajectory planning applications.

    💻 CODE 📧 MAIL
    See More Projects...