In modern game engines and robotic simulations, the demands for simulation stability and computational performance are steadily increasing. Traditional physics solvers often struggle with numerical instability and slow convergence when dealing with high stiffness problems, large timesteps, or large-scale scenarios. A recently proposed solver, Augmented Vertex Block Descent (AVBD), provides a novel solution to these challenges. AVBD not only guarantees unconditional stability, but also demonstrates superior computational performance compared to traditional methods, while achieving numerical convergence to the implicit Euler integration scheme. Its core idea is to reformulate implicit Euler integration as an optimization problem, incorporate constraints through the Augmented Lagrangian method, and treat each rigid body as an independent computational block solved in parallel on the GPU. As a general framework, AVBD can efficiently handle articulated joints, complex collisions, friction, and stacking, making it particularly suitable for large-scale, real-time simulations.
The objective of this thesis is to apply the AVBD method to an experimental digital twin platform and investigate its advantages in complex simulation scenarios. The work will involve conducting a literature review to survey state-of-the-art physics solvers, followed by the formulation of position-based constraints for multibody dynamics within the optimization framework. The student will then implement AVBD in a C++ simulation platform and exploit GPU acceleration for parallel dynamics computation. Finally, the method will be benchmarked against mainstream solvers on representative test cases to evaluate its performance, stability, and accuracy.
Keywords: Physics-based simulation, game eingine, GPU
Requirements:
- You are studying Electrical Engineering, Automation, Robotics, Computer Science, or a related field.
- You have a strong interest in game physics engines or robot simulation; ideally, you have studied multibody dynamics or robotics dynamics.
- Ideally, you have programming skills in C++.
Relevant Literature:
- Giles, C., Diaz, E. and Yuksel, C., (2025). Augmented Vertex Block Descent. ACM Transactions on Graphics (TOG), 44(4), pp.1-12.
- Chen, A.H., Liu, Z., Yang, Y. and Yuksel, C., (2024). Vertex block descent. ACM Transactions on Graphics (TOG), 43(4), pp.1-16.
Betreuer: Shao, Email:
