Joerg Stueckler

This work presents Co-VLA, which applies consensus optimization using the Alternating Direction Method of Multipliers~(ADMM) to federated VLA training, and demonstrates that Co-VLA achieves performance comparable to centralized training in both full-model training and parameter-efficient fine-tuning settings.

Object-centric Gaussian representations enable precise predictions of rigid body dynamics, even in cluttered environments with occlusions.

Differentiable forward simulations combined with guided diffusion unlock the ability to generate diverse and optimized material designs, even with discrete parameter spaces.