Hi, I'm Hokwang Choi

Project Details

This project focused on developing an advanced navigation system for autonomous robots operating in human-populated environments. The key challenge was creating a robot that could navigate through crowded spaces while respecting social norms and personal boundaries.

The system utilizes multiple stereo cameras strategically positioned to provide comprehensive 270-degree environmental awareness. These cameras work together to create a detailed 3D understanding of the surroundings, enabling the detection and tracking of dynamic objects in real-time. The stereo vision setup processes image pairs to generate accurate depth maps, which are essential for identifying obstacles, pedestrians, and other moving elements in the environment.

The pipeline integrates advanced computer vision techniques with ROS (Robot Operating System) to ensure seamless communication between perception and control modules. PyTorch was employed to implement deep learning models for object detection and classification, allowing the robot to distinguish between different types of obstacles and predict their movement patterns.

What sets this system apart is its socially-aware path planning algorithm. Rather than simply avoiding obstacles, the robot adapts its trajectory to follow social conventions - maintaining appropriate distances from people, moving predictably, and choosing paths that minimize disruption to human activities. This makes the robot more acceptable in shared spaces and improves overall safety and user experience.

credits to: Sevensense AG

Demo Video

Project Details

This project addressed a critical need in simulation-based image generation: the ability to dynamically configure simulation parameters to render different weather condtions requested by multiple clients through gRPC service. Traditional simulation workflows often require stopping and restarting simulations whenever parameter changes are needed, leading to significant time loss and workflow disruption.

The solution implements a high-performance gRPC-based communication system that exposes the Unigine simulation engine's parameters through a well-defined API.

The system architecture allows external tools, scripts, and applications to connect to the running simulation and modify parameters on-the-fly. This includes environmental conditions, sensor configurations, and scenario parameters. The C++ implementation ensures minimal performance overhead, maintaining the simulation's real-time capabilities even during parameter updates.

Demo Video

Project Details

Developed as part of the DARPA challenge, this project tackled one of the most challenging problems in autonomous systems: efficient exploration and mapping of completely unknown environments. The objective was to create an intelligent navigation system for Micro Aerial Vehicles (MAVs) that could autonomously discover and map unfamiliar spaces while ensuring safe, collision-free flight.

The system employs reinforcement learning with a carefully designed reward function that encourages the MAV to explore uncharted areas efficiently. The RL agent learns optimal exploration strategies through trial and error, balancing the need to explore unknown regions with the safety requirement of avoiding obstacles. This balance is critical in GPS-denied environments where traditional navigation methods fail.

The architecture combines deep neural networks for high-level policy learning with classical path planning algorithms for low-level control and real-time obstacle avoidance.

Even though the control function was set to ideal on hovering and moving 2D lines, the resulting system demonstrated robust autonomous behavior in complex environments, successfully mapping unknown spaces with high efficiency. The custom reward function proved crucial in guiding the agent toward exploration strategies that maximize map discovery while avoiding collisions. This work has applications in search and rescue operations, infrastructure inspection, and military reconnaissance missions where pre-existing maps are unavailable.

Research Paper

Demo Video

Project Details

Treasure Hunting is an augmented reality mobile game that blends the physical and digital worlds to create a fun gaming experience. The project demonstrates the practical application of AR technology in creating engaging, location-based entertainment that encourages outdoor exploration and physical activity.

The game leverages Android's AR capabilities to overlay digital treasures and game elements onto real-world environments as seen through the device's camera. Players explore their surroundings to discover hidden virtual treasures, creating a unique blend of gaming and real-world exploration. The AR integration ensures that virtual objects appear naturally anchored to physical spaces, maintaining proper scale.

Development utilized a combination of C++ for performance-critical AR processing and rendering, with Kotlin handling the Android application framework and user interface. This hybrid approach ensured smooth, real-time AR performance while maintaining clean, maintainable application code. The C++ components handle computationally intensive tasks like image processing, feature detection, and 3D rendering, while Kotlin manages game logic, UI, and Android system integration.

The project addressed several technical challenges including accurate spatial tracking in diverse conditions, maintaining stable AR object placement across device movements. The result is a polished mobile gaming experience that showcases the potential of AR technology in creating novel, engaging applications.

Demo Video

Project Details

This project presents a unique challenge in aerial robotics: using a quadrotor to bounce a ball repeatedly while maintaining stable flight. The objective required developing an optimal control system that could precisely manage the complex dynamics of both the quadrotor and the ball, coordinating their interactions to achieve sustained bouncing.

The core of the system is an LQR (Linear Quadratic Regulator) controller designed to compute optimal trajectories for the quadrotor. The controller must account for the full quadrotor dynamics—including thrust, torque, and rotational inertia—while simultaneously predicting ball trajectories based on physics models incorporating gravity, impact forces, and energy dissipation during bounces.

Implementation in MATLAB allowed for rapid prototyping and testing of control algorithms. The system performs real-time trajectory optimization, continuously adjusting the quadrotor's position and orientation to intercept the ball at precisely the right moment and angle. This requires solving complex optimization problems that balance multiple objectives: maintaining quadrotor stability, achieving accurate ball impacts, and minimizing energy consumption.

Research Paper

Project Details

This comprehensive aerospace engineering project encompassed the complete design, analysis, and fabrication of a human-sized rocket for a national-level scientific competition in South Korea. The project combined theoretical aerodynamics with practical manufacturing to create a functional rocket capable of stable flight.

The design process utilized Computational Fluid Dynamics (CFD) to analyze and optimize the aerodynamic properties of the rocket's key components: the nose cone, body tube, and stabilizing fins. CFD simulations modeled airflow patterns at various velocities and angles of attack, allowing the team to identify optimal geometries that minimized drag while maximizing stability. The analysis focused on critical parameters including center of pressure location, drag coefficients, and fin effectiveness across the expected flight envelope.

Stability analysis was paramount to ensure safe flight. The team conducted extensive simulations to verify that the center of pressure remained sufficiently aft of the center of gravity throughout the flight profile. This included analyzing static stability margins and dynamic response to disturbances.

Following the design phase, the team hand-manufactured the rocket body according to the optimized specifications. This involved precision fabrication of the airframe, fin mounting systems, and internal structures. The project finished with a successful launch at the national university competition, validating the CFD predictions and demonstrating the effectiveness of the integrated design approach. This hands-on experience bridged the gap between computational analysis and practical aerospace engineering, showcasing skills in fluid dynamics, structural design, and systems integration.

Research Paper

Demo Video