Robotics Researcher and Technical Leader
How can robots become more intelligent and more versatile?
This question has guided more than a decade of research and development, conducted in academic laboratories at Carnegie Mellon University, the Max‑Planck‑Institute for Intelligent Systems, and Johns Hopkins University, and in industry at Intuitive Surgical, Johnson & Johnson, Rivian & Volkswagen Group Technologies, and Autonomique. The central premise of this work is that robotic capability arises from combining a rigorous understanding of physics with well-designed interfaces between robots and objects, such as grippers and sensors.
Area of Expertise
& Versatile
Robots
Academic Labs
Industry
Focus
Research, Mentorship, and Leadership
Research & Technical Expertise
My research addresses a central question in robotics: what capabilities must a robot possess to act reliably and adaptively in unstructured physical environments? I approach this question across the full system stack, from mechanism design and sensing and actuation to feedback control and learning. A guiding premise of this work is that versatility emerges from designing these layers together rather than optimizing each one in isolation.
Mentoring & Teaching
Mentorship is an integral part of my research practice. I have mentored twelve students who have continued to graduate and professional positions at institutions including ETH Zurich, the Technical University of Munich, Delft University of Technology, Carnegie Mellon University, Johns Hopkins University, and the University of Illinois Urbana-Champaign. At Johns Hopkins University, I designed and taught a course on magnetic systems in surgical robotics. At Carnegie Mellon University, I served as a teaching assistant in mechatronic design and numerical methods.
Beyond academia, I have mentored undergraduate students through the Braven Professional Mentor Program and high-school students pursuing careers in science and engineering through the Turkish Nobel Institute.
Technical Leadership & Service
Technical leadership requires identifying the problems most worth solving and establishing the platforms and teams needed to address them. In this capacity, I founded the Advanced Robotics and Digital Solutions (A-RAD) Innovation Group at Johnson & Johnson and defined motion-control platform strategy across four automotive brands. I also serve the research community as a session chair at IROS and as a reviewer for IEEE Transactions on Robotics, IEEE Robotics and Automation Letters, Nature Communications, ICRA, and IROS.
Research
Six Research Areas, One Objective: Physical Intelligence
Each area is introduced by its central research question, followed by the institutions at which the work was conducted and selected publications.
Physical AI
Data-driven policy learning for versatile manipulation on physical robots.
Closed-loop Feedback Control
Robust feedback control under model uncertainty and delayed sensing.
Novel Sensing & Actuation
Field-based actuation and sensing: MRI, inductive, capacitive, and strain.
Medical Robotics
Untethered and clinically deployed robotic systems for minimally invasive surgery.
Mechanism Design
Physical design as a means of simplifying control and learning.
Advanced Manufacturing
Material-level programming of function and scalable experimentation.
How can robots acquire versatile skills from data and continue to improve after deployment?
Most deployed robots remain specialists, engineered for a single task in a structured environment. Learning-based approaches, including imitation learning, vision-language-action (VLA) models, and diffusion policies, offer a path toward more general behavior. On physical hardware, however, their performance is limited by factors that rarely appear in benchmarks: actuation and communication latency, contact-rich interaction, incomplete sensing, and the shift between training and deployment conditions. My research addresses this gap by combining physics-informed modeling with data-driven policy learning. This includes learned simulators for sim-to-real transfer, latency-aware inference, and human-in-the-loop correction to recover from failure modes that the demonstrations do not capture.
Institutions
Selected Publications
How can a robot maintain precision and safety when its models are inaccurate, its feedback is delayed, and its environment varies?
Closed-loop feedback control remains the foundation on which higher-level autonomy, including learned policies, is built. In practice, the difficulty lies in conditions that violate standard assumptions: low-rate or noisy state estimates, unmodeled dynamics, actuator nonlinearities, and communication delay. My work has examined these conditions across scales and applications. These include image-guided five-degree-of-freedom control of untethered magnetic robots using MRI feedback, model-predictive control of magnetic surgical needles, low-latency surgical teleoperation, and hardware-abstracted motion-control platforms for automotive systems. A recurring theme is combining physics-based models with data-driven estimation to improve robustness.
Institutions
Selected Publications
Can robots be actuated and sensed through external physical fields rather than on-board components?
At small scales, conventional motors, batteries, and cameras cannot be integrated into a robot's body. Physical fields offer an alternative. Magnetic field gradients can generate forces and torques remotely, while inductive, capacitive, and strain-based transduction can provide power and sensing without wired connections. A central part of my research showed that a clinical MRI scanner can serve both as the actuation source and as the imaging modality for an untethered robot, removing the need for dedicated external hardware. The same principle, that a robot's capabilities are bounded by its sensing and actuation modalities, motivates my current interest in tactile and force sensing for dexterous manipulation.
Institutions
Selected Publications
How can robotic systems safely access anatomical regions that conventional instruments cannot reach?
Surgical environments place strict demands on robotic systems: tissue is deformable and only partially observable, and errors are often irreversible. Two research directions are particularly relevant. Untethered magnetic robots and needles may enable minimally invasive procedures without rigid mechanical access, but they face limits in force output, localization, and control. Clinically deployed surgical platforms, in contrast, require exceptional reliability, regulatory compliance, and seamless interaction between surgeon and robot. My work spans both directions. In academic research, I have developed wireless magnetic devices for capsule endoscopy, suturing, and tissue penetration. In industry, I have worked on motion control and system integration for FDA-regulated surgical robots.
Institutions
Selected Publications
To what extent can a robot's physical design reduce the burden on its control and learning systems?
Mechanism design determines which behaviors come naturally to a robot and which must be enforced through control. A well-chosen mechanism can reduce the dimensionality of a control problem, eliminate singular configurations, or make a learning problem tractable. Examples from my research include a millirobot that achieves locomotion under a constant, unidirectional magnetic field, a condition that by itself produces no net translational force. Others are a capsule robot that reversibly locks its orientation, and nested and stacked electromagnetic coil architectures that mitigate actuation singularities. This perspective is closely related to the concept of embodied intelligence.
Institutions
Selected Publications
How can robot functionality be encoded during fabrication, and how can physical experimentation be scaled to support data-driven design?
Certain capabilities are most effectively realized at the material level. Programming magnetization profiles into soft materials enables complex, predesigned shape transformations under external fields. Miniature MRI-compatible actuators, in turn, require fabrication methods that avoid ferromagnetic components. Equally important is the scale of experimentation: automated characterization systems can execute and analyze large numbers of physical trials without supervision. These methods parallel a central need in robot learning today, where the scale and quality of physical data are primary determinants of performance.
Institutions
Selected Publications
Career
Academic and Professional Appointments
Appointments can be filtered by research area.
- 2026 –
Staff Physical AI Researcher
AutonomiquePhysical AI - 2025 – 2026
Senior Motion Control and Software Engineer
Rivian & Volkswagen Group TechnologiesClosed-loop Feedback Control - 2024 – 2025
Senior Robotics and Control Engineer
Intuitive SurgicalMedical RoboticsClosed-loop Feedback Control - 2023 – 2024
R&D Senior System Integration Engineer
Johnson & Johnson MedTechMedical RoboticsClosed-loop Feedback Control - 2021 – 2023
Postdoctoral Fellow and Instructor
Johns Hopkins UniversityPhysical AIMedical RoboticsClosed-loop Feedback ControlNovel Sensing & ActuationMechanism Design - 2021 – 2023
Robotics Consultant
Weinberg Medical PhysicsNovel Sensing & ActuationMedical Robotics - 2016 – 2020
Doctoral Researcher (Ph.D., Carnegie Mellon University, 2020)
Max-Planck-Institute for Intelligent SystemsNovel Sensing & ActuationClosed-loop Feedback ControlMedical RoboticsMechanism DesignAdvanced Manufacturing - 2014 – 2016
Graduate Research Assistant (M.Sc.)
Carnegie Mellon UniversityAdvanced ManufacturingClosed-loop Feedback ControlNovel Sensing & Actuation - 2009 – 2014
B.Sc. in Mechatronics Engineering, Valedictorian
Sabanci UniversityMechanism DesignAdvanced Manufacturing