About

Hi, I am a Computer Science master’s graduate specializing in Artificial Intelligence from Freiburg University. My current research focuses on leveraging large language models (LLMs) and vision-language models (VLMs) for industrial applications, with an emphasis on anomaly detection. I am currently an intern at Endress+Hauser, where I am working on developing a text-guided anomaly detection method that incorporates semantic insights from these models.

Previously, I explored computer vision-based approaches in robot manipulation as part of my master’s thesis in the Robot Learning Lab at Freiburg University. This work, accepted as a workshop paper at ICRA, examined 6-DoF grasp estimation of articulated objects and was supervised by Eugenio Chisari, Nick Heppert, and Prof. Abhinav Valada. As a research assistant, I contributed to computer vision datasets for tasks like object detection, pose estimation, depth estimation, and instance segmentation.

Before this, I mainly considered the theoretical aspects of Applied Mathematics. I was a research associate in the Mathematics for Uncertainty Quantification group at RWTH Aachen University, analyzing stochastic differential equations. I hold a master’s degree in Scientific Computing from Heidelberg University, where I specialized in partial differential equation analysis, and a bachelor’s degree in Applied Mathematics from Shiraz University: CV

Publication

CenterArt: Joint Shape Reconstruction and 6-DoF Grasp Estimation of Articulated Objects (ICRA Workshop)

PDF, Code, Poster, Video

• Introduce the first approach capable of jointly reconstructing 3D shapes and predicting 6-DoF grasp poses for articulated objects
• Generate a dataset of valid 6-DoF grasp poses for articulated objects
• Generate a dataset of photo-realistic kitchen scenes consisting of articulated objects

Syn-Mediverse: A Multimodal Synthetic Dataset for Intelligent Scene Understanding of Healthcare Facilities (RA-L Journal)

PDF, Website, Video

• The first hyper-realistic multimodal synthetic dataset of diverse healthcare facilities
• Provide more than 1.5M annotations spanning five different scene understanding tasks
• Provide an online evaluation benchmark along with the public dataset

Selected Projects

Policy Learning for Real-time Generative Grasp Synthesis Slides

• Design a realistic setup for mobile manipulation robot grasping in Isaac Sim
• Evaluate the performance of computer-vision-based and Policy-Learning-based approaches
• Develop an interactive imitation learning model that outperforms existing models in this setup

Robot Skill Adaptation via Soft Actor-Critic Gaussian Mixture Models Poster

• Learn a dynamical model with Gaussian mixture models from a few demonstrations
• Refine the learned Gaussian mixture model with the Soft Actor-Critic model
• Apply Autoencoder to process the input images in latent space

Optimal Importance Sampling Change of Measure for Large Sums of Random Variables Slides, Codes

• Evaluate different approaches based on Importance Sampling to estimate rare-event probabilities
• Develop an alternative change of measure using Exponential twisting that leads to the same performance
• as the optimal change of measure but without its computational limitations

Risk-Averse Optimal Control Slides

• Analysis of the underlying SDE that results in optimal strategy for Merton’s Portfolio Problem
• Study of different risk measures to consider the risk-averse version of the Portfolio problem
• Derive the solution of the SDE that describes the dynamics of risk-averse Merton’s Portfolio problem as a closed-form mathematical Formula

Analysis and Computation of Black-Scholes Equation with Local Volatility PDF

• Mathematical Analysis of the Black-Scholes equation (second order PDE) for evaluating Option contracts
• Utilize the Lagrange finite element methods to numerically solve this time-dependent PDE with mesh refinement
• To cope with shortcompings of Black-Scholes equation in real world, the local volatility function was calibrated from observed option prices in the market
• Test the obtained method on real world option prices