Projects

Below are selected research and class projects.

Causal PointNet

Refines 6D pose estimates using causal interventions and backdoor adjustments based on structural causal models. Improves robustness to viewpoint ambiguity and symmetry.

Probabilistic Digital Twin

Developed a ROS2-integrated Dynamic Bayesian Network for real-time fault detection in Universal Robots. Used Unscented Kalman Filtering to track friction, damping, and wear parameters.

Diffusion Models in Robotics

Explores how Denoising Diffusion Probabilistic Models can be used to synthesize robot action policies from noise, using a NoProp training method adapted for stochastic control.