projects

Hands-on research and engineering projects, newest first.

2025

Synthetic data curation for foundation models

Steered diffusion models to generate image-language and time-series data that reduces uncertainty in downstream predictions.

2024

Image captioning for robotic localization

Optimized in-context examples with DSPy to improve image captions for text-image robot localization.

Language-guided multi-agent reinforcement learning

Used language models to guide exploration and improve sample efficiency in multi-agent learning.

2023

Safe imitation learning for F1-Tenth

Built visual imitation learning methods with barrier-function safety constraints for autonomous driving in ROS 2.

Multi-agent search and tracking

Learned decentralized policies for teams of robots to search for and track moving targets.

Goal-conditioned quadrotor control

Studied reinforcement learning policies for waypoint reaching and trajectory tracking with quadrotors.

2022

Safe exploration for autonomous vehicles

Investigated safe reinforcement learning for vehicle navigation under state and obstacle constraints.

Multi-agent foraging

Trained cooperative agents to balance resource gathering, survival, and team goals.

Visual-inertial quadrotor navigation

Combined planning, control, and visual-inertial odometry for collision-free quadrotor flight.

Informative path planning for UAVs

Used reinforcement learning to plan aerial target discovery and classification trajectories.

2021

Traffic signal control

Developed decentralized multi-agent learning for coordination across urban traffic signals.

2020

Construction-site robot inspection

Built motion planning and mapping software for robots inspecting construction sites.

NUS Rover

Built mobility, controls, and navigation software for a student rover prototype.

Multi-robot mapping

Explored evolutionary and sampling-based planning for robots mapping unknown environments.

2019

Satellite coverage control

Developed attitude and orbit control approaches for multi-satellite coverage at Thales.

Gamifying education

Hackathon project to make middle-school learning more engaging through game mechanics.

2018

Reactive robot-arm interception

Built real-time perception, prediction, and control for a robotic arm intercepting moving objects.

Assembly with a robot arm

Used vision and force-torque feedback to assemble objects despite uncertain alignment.

Robot-arm teleoperation

Connected a haptic device, robotic arm, and soft gripper to collect expert demonstrations.

Earlier

Hexabot design

Designed and controlled a six-legged robot for a university design project.

Recurrent attention models

Explored hidden-state interpretation in recurrent models for classification.

BicycleGAN image generation

Explored multimodal image-to-image generation with BicycleGAN.

Hanoi puzzle with a robot arm

Combined computer vision and UR5 manipulation to solve the Tower of Hanoi puzzle.

SLAM and unscented Kalman filtering

Implemented particle-filter localization and mapping alongside unscented Kalman state estimation.

Warehouse state estimation

Fused odometry and camera observations to detect racks in a warehouse.

NUSSportsMatch

Student app concept for finding partners for two-player and four-player sports.