Mechanical Engineering Researcher/Controls & Robotics/UC San Diego
From airframe fabrication to flight-control integration, building aerial platforms from the hardware up.
Rigorous engineering, from first principles to flight test.
B.S. Mechanical Engineering with a Specialization in Controls and Robotics from UC San Diego, conferred September 2026 Magna Cum Laude with a 3.962 GPA and membership in the Tau Beta Pi engineering honor society.
Now reading for a research-based M.S. at UC San Diego in controls and mechatronics, continuing the aerial-robotics research in the same lab.
My background spans aerial vehicle research, embedded electronics, competitive eVTOL development, and hands-on mechanical systems. Research interests center on flight control, estimation, robust control, and robotics for autonomous and dynamic systems.
Design, development, and experimental validation of a canted-rotor hexacopter: a six-rotor aircraft that controls all six force and torque components independently, with fixed rotors and no added actuators, so it translates in any direction while holding level. The rotor cant angles come from an integer orthogonal nullspace basis of the conventional mixer, which makes the wrench matrix orthogonal and its inverse closed-form. A custom PX4 flight task bypasses the four-channel mixer and commands the full six-component wrench directly, leaving the estimator, failsafes, and ground-station tooling untouched. Across 8,094 logged samples the aircraft translates at 2.0 to 2.5 m/s² while holding under 3° of tilt, where a coplanar airframe is bound to nearly 13°.
H. Djabraian, D. Mkrtchian, T. Wan, and T. R. Bewley, “Direct 6-DoF Control of a Canted-Rotor Hexacopter with No Added Actuators,” submitted to the IEEE International Conference on Robotics and Automation (ICRA), Seoul, 2027. Under review.
Sole teaching assistant for the department's hands-on embedded controls and robotics course, 61 students, run by the same lab as the hexacopter research above. The course builds real control hardware, so the support load is lab bring-up and debugging rather than problem sets.
Wrote the grading rubrics and graded homework sets and a midterm for a 55-student classical control course.
Led coordination across four engineering teams (Airframe, Embedded Controls, Software/Automation, and Proposal) developing a wildfire-response eVTOL for the 2025–2026 Vertical Flight Society Design Competition. Directed the Embedded Controls team in design and analysis of the aircraft's power and control systems (batteries, PDB, ESCs, flight controller, sensors). Owned technical documentation including the Safety and Risk Plan, Wiring Plan, and Final Technical Report.
Diagnosed and repaired mechanical systems across a wide range of vehicles, from carbureted engines to modern computer-controlled drivetrains. Serviced and overhauled powertrain, suspension, and braking systems on high-performance vehicles to strict engineering tolerances. Introduced data-driven diagnostics using OBD-II and digital repair repositories. Prepared technical reports and warranty claims, securing over $11,000 in reimbursements.
Designed and fabricated a 3D-printed chute-and-tubing system to capture and redirect overflow fluid from a pressurized jet apparatus used in fluid flow and heat transfer research. Iterated CAD models to optimize fit, flow routing, and durability. Final design successfully contained excess fluid, enabled recovery for reuse, and prevented equipment damage. Designed to be adaptable for closed-loop fluid reservoir integration.
Designed and built an RFID-controlled dual-guillotine gate airlock system connecting a group-housing cage to an existing touchscreen training chamber. Passive RFID transponders implanted in each mouse allow the system to identify individual animals and enforce single-occupancy access automatically, eliminating manual handling and individual housing. Reduced cohort training time from 5 to 6 months down to 10 to 15 days. Measured ~430 ms from tag detection to gate fully open against a 500 ms requirement, and supports 4 or more mice per cage simultaneously. Delivered to the lab for use.
A 1/10-scale autonomous vehicle that takes spoken commands, follows a person using stereo depth, and brings its own autonomy stack up at power-on. Compute is split off the vehicle across three machines: speech recognition in a phone web app, language parsing on a remote GPU across a private mesh network at a measured 35 ms round trip, and object detection on the camera's own vision processor. The person detector was trained without hand-labelled data, using an open-vocabulary teacher to auto-label 5,000 images and distilling into a YOLO11n student at mAP50 0.816, deployed to a Myriad X VPU at 25.6 fps. The ROS 2 stack is arranged so exactly one node publishes to /cmd_vel, arbitrating voice, autonomy, and safety by priority behind a killswitch that bypasses speech recognition entirely.
A real-time visual servoing platform: a two-axis pan-tilt turret that detects, estimates, and follows moving targets using computer vision and closed-loop control. The camera sits on the moving axes, so the sensor and the plant are coupled and the loop has to lead the target rather than chase it. Built end to end: mechanism, firmware, vision pipeline, state estimation, control law, and a browser control surface. The plant was identified by step-response sweep and LQR and LQG designed against it, running alongside PID on the hardware.
Next objective: autonomous lock-on and tracking of the canted-rotor hexacopter in flight, closing the loop between both research platforms.
«Ἦθος ἀνθρώπῳ δαίμων»
"Character is destiny."
Heraclitus
Open to research collaborations, graduate opportunities, and engineering conversations. Reach out directly or grab the latest CV.
Download CV (PDF)