1. Theses & HiWi Jobs
    2.  

Physics-Informed Trajectory Sampler for Autonomous Driving

Institute
Professur für autonome Fahrzeugsysteme
Type
Semester Thesis /
Content
 
Description

Modern autonomous driving systems often rely on data-driven trajectory planners that learn motion patterns from expert demonstrations or simulation rollouts. While effective in many scenarios, these neural samplers often overlook physical feasibility—resulting in kinematically implausible or dynamically unstable trajectories. To address this, we propose a physics-informed trajectory sampler that incorporates vehicle dynamics directly into the loss function, leveraging the computational power of GPUs for efficient processing.

At the AVS Lab, we are developing a neural network-based trajectory sampler that integrates vehicle dynamics constraints, ensuring that generated trajectories are both physically feasible and dynamically stable. By embedding physics into the learning process, our approach bridges the gap between data-driven methods and traditional kinematics-based planners, which typically run on CPUs and lack the speed and flexibility of GPU-accelerated neural networks.

Project Objective:

This project aims to enhance trajectory planning for autonomous vehicles by integrating vehicle dynamics into neural network-based samplers. Core components include:

  • Designing a loss function that incorporates vehicle dynamics constraints
  • Implementing a GPU-accelerated neural network for real-time trajectory sampling
  • Evaluating the sampler's performance in terms of physical feasibility and computational efficiency
  • Comparing the proposed method with traditional kinematics-based planners

We offer:

  • Cutting-edge research at the intersection of physics and machine learning
  • Integration with existing autonomous driving systems
  • Opportunities to publish in top-tier robotics/AI venues
  • English-speaking environment, with thesis supervision possible

Your qualifications:

  • Strong Python and PyTorch skills
  • Interest in autonomous driving and deep learning
  • Background in vehicle dynamics or physics-informed machine learning is a plus

Start date is flexible. If you're interested, please send:

  1. A highlight performance description of your computer science (e.g., deep learning, coding) related courses (can be within campus or online) (200 words would be enough)
  2. Academic transcript (optional for now, but may be required later)
  3. CV (optional)

to dingrui.wangtum.de.

Possible start
sofort
Contact
Dingrui Wang
dingrui.wangtum.de