About Me
I am a PhD candidate, expected to graduate in 2026, with experience and interest in artificial intelligence for robotics and autonomous vehicles. My research focuses on developing AI systems for autonomous vehicles in mixed-autonomy traffic and safety and efficiency verification of automated vehicles in simulation. I have hands-on experience with multi-agent reinforcement learning, foundation models, and world models for learning and verification, especially in complex and uncertain environments.
News
- Sep. 20, 2026Our paper NOMAD has been accepted to NeurIPS 2026! Check out the project page and the preprint on arXiv.
- Jun. 17, 2026Our paper DiscoGen has been accepted to ICML 2026 in Seoul, South Korea! Read the preprint on arXiv.
- May. 20, 2026I will be giving a talk at Eindhoven University of Technology about self-play RL for autonomous driving.
- Apr. 1, 2026Gave a talk at Waymo UK about NOMAD.
- Mar. 18, 2026New preprint: DiscoGen, a procedural generator of algorithm discovery tasks for machine learning. arXiv.
- Mar. 15, 2026Made the code of NOMAD public. Check it out on GitHub.
- Mar. 10, 2026Gave a talk at Motional about NOMAD.
Experience
- Co-developed PlanNudge, a lightweight RL nudge policy that corrects a frozen diffusion-based driving planner, matching or exceeding the previous state-of-the-art RL finetuning method on all six nuPlan benchmarks with ~6x fewer environment steps (submitted to CVPR 2027)
- Co-developed a Self-play MARL framework for improving the performance of agents transferred from one environment to another, achieving 80-120% success rate improvement (NeurIPS 2026)
- Co-developed DiscoGen, a procedural generator of algorithm discovery tasks across ML domains including RL, vision, language modeling, and trajectory prediction (ICML 2026)
- Implemented behavior cloning (BC) policies in complex and uncertain environments with more than 60 agents in each scenario
- Training transformer-based vision models for predicting abnormal scenarios from dashcam videos
- Conducting a comparative analysis of rule-based vs. AI-based anomaly detection methods in autonomous systems
- Contributed to the development of Hi-Drive automated driving function (ADF), a simulation model for longitudinal and lateral control of sim agents in both motorways and urban areas
- Led the working group on simulation parameterization for virtual verification of AVs; calibrated 30+ parameters
- Developed methods to extract edge cases and safety-critical scenarios from Hi-Drive, Waymo, and Lyft datasets
- Collaborated with a team of 50+ people from 30+ car manufacturers and institutions in Europe; 7 deliverables
- Developed an award-winning real-time MPC-based simulation framework for autonomous vehicles (IEEE ITSC 2023 Award Winner)
- Leveraged Foundation Models (VLMs) for critical trajectory failure detection with 90%+ accuracy on zero-shot learning
- Developed a hybrid MPC-RL framework for motion planning with 50% higher efficiency
- Co-designed simAgent algorithms for the EU Horizon 2020 flagship Project Hi-Drive
Publications
View full publication list
Google Scholar profile
Projects
NOMAD: Driving in New Cities Without Demos
RL and imitation learning framework for training autonomous driving agents that generalize to new cities without human demonstrations.
DiscoGen: Algorithm Discovery Benchmark
Procedural generator of algorithm discovery tasks across ML domains including RL, vision, language modeling, and trajectory prediction.
Heterogeneous Multi-Agent Driving
VAE-based heterogeneous agent modeling in PufferDrive for realistic multi-agent driving simulation with diverse behavior styles.
MPC*RL: Hybrid Planning for AVs
Integrating Model Predictive Control with Deep RL for safe and efficient autonomous vehicle control at unsignalized intersections.
Simulation Framework for Mixed-Autonomy Traffic
Real-time simulation framework for evaluating autonomous vehicles in mixed traffic scenarios with human-driven vehicles.
Human-AV Interaction at Intersections
Data-driven analysis of autonomous and human-driven vehicle interactions at unsignalized intersections using Waymo and Lyft datasets.
Interactive MPC Learning Tool
Educational GUI application for exploring Model Predictive Control with interactive trajectory design and real-time parameter tuning.
Imitation Learning for AVs Inside World Models
RNN-based World Model for training autonomous vehicles through imitation learning in complex driving scenarios.
Edge Case Detection for Autonomous Driving
Foundation model-based system for detecting critical edge cases in autonomous driving systems (Hi-Drive Project).
Education
hover or tap to expand
- Advisors: Prof. Bart van Arem, Dr. Simeon C. Calvert
- IES IROS SYPA Award 2025
- TU Delft TMI Grant for Collaborative Research 2025
- Research Topic: Multi-agent self-play reinforcement learning for generalizable agents
- Advisors: Prof. Shimon Whiteson, Prof. Jakob Foerster
- GPA: 3.94/4.00
- Graduate Student Dean's Award
- Elite Graduate Student Award – Iran's National Elites Foundation
- GPA: 3.67/4.00