Robotics systems
ROS workflows, real-time control, and collaborative task design on Tiago (PAL Robotics) and the Furhat social robot. End-to-end systems that run live, not just in simulation.
Robotics & AI Engineer · Human–Robot Interaction
I build robots that detect, communicate, and recover from failure.
Ph.D. researcher and full-stack robotics engineer. I design end-to-end systems on real robots — from ROS control and perception pipelines to LLM/VLM-based reasoning — and validate them through real-world experiments with 50+ users.
- ROS
- Python
- LLMs / VLMs
- Computer Vision
- NVIDIA Isaac
- Behavior Trees
- Unity
- Statistical Modeling
50+ participants in real-robot user studies
4 robot platforms shipped on (Tiago, Furhat, Nao, Opo)
5+ peer-reviewed publications (CHI, HRI)
#1 Winner, HRI'24 Robot Challenge
What I do
Engineering strengths
I work across the full stack of a robotics system — hardware integration, perception, reasoning, and the human studies that prove it works.
Perception & data
Eye tracking with Pupil Labs, OpenCV computer-vision pipelines, ROS synchronization, and real-time feature extraction. Built annotation workflows for multimodal interaction data.
AI reasoning
LLM- and VLM-based reasoning via OpenAI APIs, integrated with behavior trees in NVIDIA Isaac for autonomous failure detection and recovery. Python throughout.
Studies & analysis
Experimental design, study execution with 50+ participants, and rigorous analysis — linear mixed-effects and cumulative link mixed models — to turn behavior into evidence.
Selected work
Projects & systems
A sample of systems I designed, built, and evaluated on real robots.
Gaze-based failure detection
A pipeline that fuses eye tracking, ROS synchronization, and real-time gaze features to detect robot failures from human reactions — with a random-forest classifier flagging failures within seconds.
- Pupil Labs
- ROS
- OpenCV
- Python
Autonomous failure recovery
A simulation-based failure detection framework in NVIDIA Isaac combining behavior trees with VLM-based reasoning, enabling a robot to autonomously detect and recover from errors.
- NVIDIA Isaac
- Behavior Trees
- VLM
- Python
Trust under repeated failure
A collaborative Tangram system on Tiago that scripts controlled robot failures and awareness behaviors, used in a 54-participant study on how trust evolves across repeated errors.
- Tiago
- ROS
- Study design
- Mixed models
Furhat emotional expressions
Complex emotional expressions built from facial action units on the Furhat robot, evaluated with both VLM-based and user-based methods — and demoed live for public audiences.
- Furhat
- FACS
- VLM eval
OfficeMate assistant on Tiago
An office-assistant robot with autonomous navigation and interaction on Tiago — winner of the HRI'24 Robot Challenge with Team Melbourne.
- Tiago
- Navigation
- ROS
ROSAnnotator
A web application for analyzing ROSBag data in human-robot interaction, making multimodal interaction logs searchable and annotatable. Supervised through to delivery.
- Web app
- ROSBag
- Annotation
Track record
Experience
Dec 2023 – Present
Researcher & Engineer — University of Melbourne
Interactive Technologies Lab (IXT) · Melbourne, Australia
- Built ROS-based collaborative tasks with diverse failure conditions and ran controlled studies with 50+ participants on real robots.
- Developed pipelines integrating eye tracking, ROS synchronization, and real-time gaze feature extraction.
- Designed a simulation-based failure detection framework in NVIDIA Isaac using behavior trees and VLM-based reasoning.
- Delivered live robotics showcases and public demos.
Sep 2021 – Sep 2023
Research Assistant — Sharif University of Technology
CEDRA · Tehran, Iran
- Programmed and ran HRI experiments with Nao and Opo robots studying gaze behavior.
- Built deep-learning models for lip reading and facial emotion recognition from video.
Let’s build something that works in the real world.
I’m open to robotics, ML, and AI engineering roles. Happy to talk through any of the projects above.