Safe learning and control for autonomous systems, with formal constraint-satisfaction and safety guarantees

• Constrained neural networks and hard constraint satisfaction
• Safe-by-design neural network controllers and control barrier functions
• Model predictive control, reachability analysis, and safety-critical autonomy

Safe autonomy, motion planning, and control for robotic systems

• Verifiable perception-control architectures and safety assurance for autonomous vehicles
• Robust collision avoidance and safe operation in cluttered environments
• Optimization-based motion planning with B-spline continuous-time reference generation for robotic platforms

Noise Effects on Low-Resolution Face Recognition Algorithms

• Implemented Identity-Aware LRFR on synthetic and real low-resolution images.
• Compared recognition accuracy across synthetic and real inputs.
• Demonstrated the performance gap between idealized and real environments.
• Evaluated how Gaussian, Poisson, and pepper-and-salt noise affect recognition performance.

Low-Resolution Face Recognition Algorithms and Noise Effects

• Implemented face detection and alignment with Multi-task Cascaded Convolutional Networks on FERET and SCface.
• Compared Identity-Aware LRFR with SphereFace for low-resolution face recognition.
• Proposed improvements to make LRFR methods more robust to noise in surveillance-style settings.

• Worked on a paper-based LED screen embedded in clothing.
• Worked on a remotely controlled robot arm based on a piezo sensor.

• Independently developed time sequence forecasting models for automated-store customer demand under changing weather, traffic, and time conditions.
• Tested Vector Autoregression, Long Short-Term Memory, and Facebook Prophet models, and demonstrated their use on an air-pollution forecasting dataset.
• Annotated images with labels for head, hand, and position using makesense.ai to support YOLO-based human-detection models.

• Evaluated isotropic and anisotropic friction effects between the swimmer body and ground in the MuJoCo simulator.
• Developed reinforcement-learning control algorithms to move the snake in a desired direction on flat ground.

• Developed an intelligent turnstile system suitable for both left-handed and right-handed users.
• Integrated Arduino with a ToF sensor to detect gesture-based distance patterns and respond correctly to both left- and right-hand opening motions.

• Developed and presented an intelligent duct-cleaning robot proposal and business plan.

• Led a team of eight volunteers to provide teaching services at a local primary school.

• Helped construct a treehouse and wooden bridge with around 30 volunteers to improve local tourism and transportation.

• Promoted volunteer teaching activities with other committee members.
• Scheduled interviews for around 200 applicants and supervised the recruiting process to select qualified volunteers for underserved regions.

• Provided volunteer teaching services to local primary- and high-school students.