[ 00 ] ● SYSTEM ONLINE — LIDAR SWEEP ACTIVE

KARIM
HAMMOUD

Engineering machines that walk, fly, sense and learn.

> ROBOTICS_

[ MSC ]UCL · Distinction

[ BENG ]AUB · GPA 3.83

[ FOCUS ]Humanoids · RL · Teleop

[ 01 ] PROFILE

I design, simulate and control robotic systems — from reinforcement-learning policies that teach humanoids to walk, to the sensors, firmware and models that let machines understand the world around them.

0%[ DETECTION ACCURACY ]MedAssist IoT tracker

0[ BILATERAL ARMS ]Franka Emika Panda ×2

0[ PAYLOAD ]Agricultural octocopter

0[ NATURAL FREQ ]MEMS gyroscope

[ 02 ] RESEARCH

Mission files

R-01 / HARDWARE + DATA PIPELINE

Egocentric capture headset & processing pipeline

Designed a head-worn, four-camera capture headset that records egocentric data from human workers as they carry out real tasks — a first-person record of what their hands and bodies are doing, captured where the work actually happens.

Recordings then run through a post-processing pipeline: PII scrubbing removes faces and identifying details, 3D pose estimation recovers body and hand motion, depth estimation adds per-frame scene geometry alongside the synchronised IMU streams, and an annotation pipeline labels it all into a training-ready dataset.

Hardware Design
Computer Vision
3D Pose
Depth Estimation
Data Pipelines

Four-camera egocentric capture headset in white and orange — 4 × CAM · EGOCENTRIC CAPTURE

Post-processing viewer showing the RGB feed with privacy blur, depth map, IMU plots and 3D pose in a point-cloud scene — POST-PROCESS · PII ▸ POSE ▸ DEPTH ▸ LABEL

R-02 / INDUSTRY ROLE

Humanoid locomotion through imitation-driven RL

Developed and deployed dynamic, robust reinforcement-learning policies for proprietary and commercial humanoid robots, enabling complex locomotion across varied environments.

Instead of hand-tuned rewards, a modified imitation-learning curriculum consumed kinematically feasible motion data captured in a VR simulation, where human operators intuitively guided a kinematically controlled robot. The result: robots that walk, balance and adapt — closing the gap between simulation and the real world.

Reinforcement Learning
Imitation Learning
Humanoids
Control

Render of a humanoid robot mid-stride — SIM ▸ REAL · POLICY v.FINAL

R-03 / MSc THESIS

4-channel bilateral teleoperation with learned force

Built a 4-channel bilateral control system across two 7-DoF Franka Emika Pandas, letting two people perform robot-mediated, human-to-human tasks. Controllers in C++, ROS and Python switch between Cartesian and joint references depending on the motion transferred.

A PatchTST transformer was trained by imitation learning to generate human-like force profiles, and evaluated with a modified Turing test — it convincingly replicated human force dynamics.

C++
ROS
Python
MATLAB
Transformers

Read full thesis ↗

Render of two Franka arms linked in bilateral teleoperation — MASTER ⇄ SLAVE · 1 kHz

R-04 / PAPER

Mobile Franka, third-person VR

Enabled manipulation of a mobile Franka arm from third-person VR. Joystick control drives the mobile base to location; a Gaussian splat then reconstructs the scene for real-time, sim-to-real third-person manipulation of the arm.

C++
ROS
Python
Gaussian Splatting
VR

Read the paper ↗

Render of a Franka arm on a mobile base with a Gaussian-splat scene reconstruction — SPLAT RECON · LIVE

R-05 / BEng FINAL YEAR PROJECT

Autonomous agricultural octocopter

Designed and built a low-cost octocopter for pesticide spraying: automated path optimisation between user-defined waypoints, and a return-to-base routine for refilling. Navigation and control run on MAVLink; modular, lightweight construction carries a 4 kg payload.

Expand flight log

Simulation. ANSYS FEA evaluated structural integrity under operational stress. Modal analysis kept the airframe clear of damaging resonance; dynamic loading showed minimal deformation at maximum payload and thrust.

Testing. Field trials covered flight stability, propulsion and payload distribution. Manual and GPS-based autonomous path planning delivered precise, repeatable spray patterns, and simulated motor failures were absorbed by redistributing power for controlled landings.

Hardware. Composite airframe, high-efficiency brushless motors, a flight controller fusing GPS and IMU, a 5 L tank with a custom brushless pump and calibrated spray array, and carbon-fibre landing gear.

V2. Carbon-composite structure, CFD-refined aerodynamics and modular payload bays for broader agricultural and industrial use.

SolidWorks
MAVLink
ROS
ANSYS

Render of the agricultural octocopter spraying a field — PAYLOAD 4 KG · RTB ENABLED

[ 03 ] ARCHIVE

Projects

[ 04 ] TRAINING DATA

Education

UCL · LONDON, UK

MSc Systems Engineering for the Internet of Things

▲ Graduated with Distinction

Embedded Systems Design
Sensor Systems
Real-world Multi-agent Systems
Robotic Control Theory
Speculative Design

AUB · BEIRUT, LEBANON

BEng Mechanical Engineering, Minor in Bioengineering Design

▲ GPA 3.83 / 4.00 · Cum Laude · Dean's Honor List

Control Systems
MEMS
Optimization
Thermodynamics
Mechanics of Machines
CAD/CAM
Heat Transfer
Biomaterials & Medical Devices
Tissue Engineering

[ LANGUAGES ] EnglishArabicFrenchSpanishPortuguese

[ 05 ] OPEN CHANNEL

Get intouch↗

Karimhammoud1@outlook.com LinkedIn ↗

KARIMHAMMOUD