Seokhyun Kim

Seokhyun Kim

I am an M.S. student in Mechanical Engineering at Carnegie Mellon University who designs, builds, and controls robots. My work spans mechanical design, electronics, sensing, simulation, and control, from a 33 g untethered quadruped to a cat-inspired robot that rights itself in midair.

Research Projects Contact
Portrait of Seokhyun Kim

About

I enjoy taking robotic systems from an idea to a working prototype.

I am a graduate researcher in the Microrobotics Lab and the Biorobotics Lab at Carnegie Mellon University, and I previously studied decentralized locomotion control in the Real-World Computing Lab at Tohoku University. Across these labs, I have built legged, underwater, and aerial robots inspired by animals.

I do my best work where hardware and software meet: designing the mechanism, building the electronics, and writing the controller so the whole robot works reliably. I am looking for robotics research and development roles where I can take systems from an idea to a tested prototype.

Research

Research across legged, underwater, aerial, and bio-inspired robotic systems.

Lil-Quad robot next to a 9-volt battery for scale.

Current-based proprioception · Small-scale locomotion

Lil-Quad

Small legged robots could search collapsed buildings or inspect tight spaces, but at the centimeter scale most of them run open-loop, because there is little room for sensors. My project, Lil-Quad, set out to give a robot this small real feedback by using the one signal every motor already produces: its current. I designed and built an untethered, RHex-inspired quadruped about the height of a 9-volt battery. It has an FDM-printed ABS torso, SLA-printed legs that combine rigid and compliant materials, and a custom PCB with a microcontroller, motor drivers and current sensors. From those current signals, a neural network I trained detects when each leg touches down and lifts off, so the robot senses ground contact without encoders or foot sensors. I'm now building on that sensing to make the gait more stable. That means refining the PD controller, redesigning the legs to keep the front feet on the ground, and exploring learning-based control.

Motor current sensing PCB design Embedded systems Machine learning 3D printing
HUMRS underwater snake robot undergoing testing in a pool.

Underwater robotics · State estimation

HUMRS

Underwater infrastructure such as ship hulls, pipes and dams needs regular inspection, but divers are costly and bulky vehicles can't reach tight spaces. HUMRS (Hardened Underwater Modular Robot Snake) is an articulated snake robot that can slip into those spaces. It already swims well under teleoperation, but a human operator still has to guide every move. To give it more autonomy, the robot first needs to see and understand where it is, and that is hard underwater. Light fades, the water is murky and there are few visual features to track. I'm working on improving the robot's underwater vision. I integrate its cameras and sensors in ROS and adapt VINS-Fusion, a visual-inertial method, to estimate the robot's position and orientation from camera and IMU data. I'm also working on waterproof mechanical design so the vision hardware works reliably during pool tests. Better state estimation is the step that will let HUMRS move beyond teleoperation toward autonomous inspection.

ROS VINS-Fusion Vision-based state estimation Waterproof design
Cat-inspired aerial self-righting robot with an actuated spine and tail.

Aerial self-righting · Bio-inspired robotics

Cat-Inspired Robot

When a cat falls, it twists its spine and swings its tail to land on its feet. Robots rarely have that ability, so an unexpected fall can expose their body and electronics to a damaging impact. This project builds a cat-inspired robot to study how a body reorients in midair, and whether an actuated tail helps. It combines a 2-DoF articulated spine with a 1-DoF actuated tail, and onboard sensors feed a reinforcement learning controller trained in MuJoCo. My main role was the robot's mechanical and electrical design: the structure, spine, tail, onboard electronics and wiring. I also helped with the control side. In drop tests, the robot recovered to a feet-down orientation within 2 m of free fall, even when released upside down. With the tail, it needed only 1 m and recovered more completely, at the cost of 8.8% more mass.

Mechanical design Electrical design Reinforcement learning MuJoCo
Simulation of whole-body quadruped coordination with limb linkages and contact markers.

Decentralized coordination · Bio-inspired control

Whole-Body Quadruped Coordination

Animals don't plan every footstep. Their legs respond to the forces they feel, and coordination emerges from those local responses. Most legged robots instead follow precomputed foot trajectories, which can break down on terrain the designer didn't plan for. This project investigates how coordination between the legs, and between the body and the legs, can emerge from decentralized control. Each joint is driven by a central pattern generator (CPG), and ground reaction force (GRF) feedback adjusts its rhythm based on the load it feels, so no full foot trajectory is ever prescribed. Building on existing CPG controllers for the hips and shoulders, I built a quadruped simulation with a flexible body and designed the CPG and feedback rules for the knee and elbow joints. With feedback control, the simulated quadruped produced a lateral-sequence walk, the walking gait seen in many four-legged animals, without that gait being programmed in.

CPG GRF feedback Simulation Bio-inspired control

Projects

Hands-on robotics and mechatronics projects outside the research section.

E-DUCATI-ON self-balancing motorcycle with a reaction wheel.

Mechatronics · Engineering education

E-DUCATI-ON: Self-Balancing Motorcycle

Feedback control is hard to learn from equations alone, because students rarely see a controller act on real hardware. E-DUCATI-ON is a self-balancing motorcycle built as a teaching platform for PD control. A reaction wheel spins up to produce corrective torque that keeps the motorcycle upright. Through a simple control interface, students adjust the proportional and derivative gains and immediately see how the motorcycle responds. Too little damping makes it oscillate, and too much makes it sluggish. I designed and built the platform, including its 3D-printed frame, electronics, and balancing controller.

PD control Mechatronics 3D printing Embedded systems
Prototype volumetric display hardware with a dense LED array and supporting electronics.

Personal project · Embedded electronics

Volumetric Display

I built this volumetric display for fun, to see whether a flat LED panel could draw images that appear three-dimensional. It combines a dense LED array, a custom 3D-printed structure, and embedded electronics that control each LED.

LED arrays Embedded electronics 3D printing

CV

Academic background & technical skills

View CV Download CV

Education

2025–present Carnegie Mellon University

M.S. in Mechanical Engineering

2021–2025 Tohoku University

B.Eng. in Electrical, Information and Physics Engineering

Technical skills

  • Python, MATLAB, C/C++, and embedded firmware
  • ROS, MuJoCo, and simulation-based control development
  • Machine learning model training and deployment
  • VINS-Fusion and vision-based state estimation
  • CAD, 3D printing, electronics integration, and PCB-related hardware work

Professional profile

Additional background, experience, and updates are available on LinkedIn.

Download CV View LinkedIn

Contact

Let’s connect

If you would like to talk about robotics research, collaboration, or related opportunities, feel free to reach out.

hyun10028@gmail.com LinkedIn