A curated collection spanning perception, controls, learning, and systems.
🧠
UIUC · Robot Learning
VLA Model for Language-Guided Humanoid Loco-Manipulation
CLIP-based VLA with Teacher–Student distillation via DAgger. EfficientNet/BERT student deployed to edge hardware. 62.1% success on unseen tasks.
CLIPVLABehavior CloningDAggerEfficientNetBERT
Oct 2025 – Dec 2025
Language-Guided Humanoid Loco-Manipulation via Vision-Language-Action Models
UIUC Graduate · Robot Learning
Architected a Teacher-Student distillation pipeline, training a lightweight student policy to clone behaviors from a CLIP-based HumanVLA teacher via the DAgger algorithm to resolve distribution shift.
Optimized inference latency by deploying the EfficientNet/BERT-based student policy to edge hardware, achieving real-time control rates (62.1% success on unseen tasks).
Evaluated generalization against visual domain randomization, demonstrating superior robustness compared to baseline non-distilled policies.
AutoShield — Real-Time Pedestrian-Intent Prediction with Safety-Filtered Autonomous Driving
UIUC Graduate · Perception & Controls
Implemented a multi-modal perception stack by performing extrinsic calibration and projection of LiDAR point clouds onto RGB-D depth maps to associate geometric obstacles with semantic classes.
Engineered a Stanley Controller for lateral control, tuning cross-track error and heading error gains to ensure stability at higher velocities on the GEM e4 hardware platform.
Developed a probabilistic safety layer that computes Time-to-Collision (TTC) using relative velocity vectors, filtering pedestrian motion noise via a Kalman Filter to minimize false positives in emergency braking.
Evaluated planner behavior under adversarial pedestrian crossings and partial sensor dropout, demonstrating robust fail-safe engagement with 91% success across various real-world trials.
Hierarchical Multi-Agent RL for Humanoid Interaction
Hierarchical PPO for Unitree G1 humanoids: high-level navigation decoupled from 29-DOF locomotion. Curriculum learning in Isaac Lab with domain randomization.
Hierarchical Multi-Agent Reinforcement Learning for Humanoid Robot Interaction
UIUC Graduate · Robot Learning & Controls
Designed a hierarchical control architecture for Sim-Unitree G1 humanoids, decoupling high-level PPO navigation policies from low-level PPO policies governing active 29-DOF joint actuation.
Engineered a curriculum learning schedule and height-scan state observations to train robust locomotion gaits capable of traversing jagged, uneven terrains in Isaac Lab simulations.
Open-vocabulary 6D pose tracking extending FoundationPose with Moondream2 VLM + SAM-3 segmentation. 88.31% ADD-S on 44 YCB-Video datasets.
FoundationPoseVLMSAM-3TripoSRPyTorchZero-Shot
Oct 2025 – Dec 2025
Open-World Zero-Shot 6D Pose Estimation
UIUC Graduate · Perception
Architected a 6D pose tracking pipeline extending FoundationPose to handle open-vocabulary objects by integrating Moondream2 for semantic retrieval and SAM-3 for instance segmentation.
Synthesized geometric proxies by building a mesh retrieval system that queries Objaverse-XL and generates assets via TripoSR, enabling zero-shot generalization to unseen objects.
Engineered a relabeling loop using Gemini VLM to automatically correct semantic drift during long-horizon tracking sequences.
Gemini VLM task planner + ROS2 Nav2 on Booster K1 Humanoid synced with Snap AR Spectacles via WebSocket at sub-100ms latency.
ROS2Gemini VLMNav2OpenCVWebSocketAR
Oct 2025 - Oct 2025
C.A.R.E. — Companion Autonomous Robotic Entity
CalHacks 12.0 · Systems & Perception
Developed an embodied AI stack integrating a Gemini VLM-driven task planner with a ROS2 Navigation2 backend for semantic goal navigation on the Booster K1 Humanoid.
Engineered a WebSocket bridge to synchronize state between Snap AR Spectacles and the robot, enabling low-latency heads-up display interactions for human-robot collaboration.
ROS2,Nav2,Navigation Stack,Gemini VLM,OpenCV,Booster K1,Snap AR Spectacles,WebSocket,Human Detection,Object Tracking,Task Planning,Teleoperation,Real-Time Systems,Python
Curved Glass Cleaning Robot for Changi Airport Group
SUTD Capstone · Systems, Controls & Hardware
Led 7-member team to design and deploy an AGV capable of reaching 2m height and cleaning curved glass surfaces.
Architected distributed control stack using ROS2 on NVIDIA Jetson Orin Nano with dual Arduino microcontrollers via Finite State Machine.
Engineered cascade lift mechanism hoisting 15kg payload to 2m in 5s with closed-loop encoder feedback.
Designed custom force-current-based admittance controller for glass contact without damage.
Validated in live airport environment with magnetic tape navigation and WebSocket-based GUI.
ROS2,NVIDIA Jetson Orin Nano,Admittance Control,Finite State Machine,Arduino,Encoder Feedback,Closed-Loop Control,Path Planning,WebSocket,Real-Time Systems,SolidWorks,Embedded Systems,Linux,Team Leadership
glass-cleaning-robot.mp4
⛈️
SUTD · Deep Learning
Category 1 Lightning Risk Prediction using CNN-RNN
CNN-RNN model predicting Cat 1 Lightning Risk from weather station features: rainfall, wind speed, temperature, humidity, wind direction.
CNN-RNNTime SeriesPyTorchWeather
Jan 2025 - Apr 2025
Category 1 Lightning Risk Prediction using CNN-RNN
SUTD Coursework · Deep Learning
Developed CNN-RNN model to predict Category 1 Lightning Risk warnings trained on features such as rainfall, wind speed, air temperature, relative humidity, and wind direction from weather stations.
CNN,RNN,Time Series Prediction,PyTorch,Deep Learning,Weather Data,Feature Engineering
🏎️
SUTD Thesis · SHARP
DIAPP — Dynamics-Intent Aware Pure Pursuit Controller
Non-linear tire slip + friction circle vehicle model with modified Pure Pursuit for high-lateral-acceleration cornering. Significant CTE reduction.
Pure PursuitVehicle DynamicsTire Slip ModelPythonMATLAB
Sep 2024 – Apr 2025
Novel Dynamics-Intent Aware Pure Pursuit Controller (DIAPP)
SUTD SHARP Thesis · Controls
Derived a dynamic vehicle model incorporating non-linear tire slip angles and friction circle constraints to predict trajectory deviation under high-lateral-acceleration cornering.
Engineered a modified Pure Pursuit control law that dynamically optimizes curvature selection by compensating for estimated slip vectors.
Benchmarked controller in simulation, demonstrating significant reduction in cross-track error during aggressive maneuvering.
Pure Pursuit,Vehicle Dynamics,Tire Slip Modeling,Friction Circle,Cross-Track Error,Lateral Control,Trajectory Tracking,Path Planning,Python,MATLAB,Simulation
🏁
SUTD Coursework
Lane-Tracking 4WD Race Car
Vision pipeline on Jetson with adaptive thresholding + radial-scan. Real-time polynomial lateral controller at 45+ FPS.
JetsonOpenCVLane DetectionReal-Time45+ FPS
Jan 2025 – Apr 2025
Lane-Tracking and Object Detection of Outdoor Scaled 4WD Race Car
SUTD Coursework · Perception & Controls
Optimized vision-based lane perception on NVIDIA Jetson with adaptive thresholding and radial-scan for variable lighting.
Implemented real-time lateral controller via polynomial curve fitting at 45+ FPS.
Pure Physics and Chemistry · Fundamentals of Electronics (Applied Subject)
Research
Publications
Peer-reviewed contributions to robotics and biomedical AI.
19th IEEE International Conference on Automation Science and Engineering (CASE 2023)
Evaluating Visual Odometry Methods for Autonomous Driving in Rain
Benchmarked classical and learning-based VO/SLAM across Oxford RobotCar, 4Seasons, and internal Singapore rain datasets. Proposed DROID-SLAM heuristic variant with map priors for rain robustness.
Future Generation Computer Systems (FGCS), Elsevier, 2019
Characterization of Focal EEG Signals: A Review
Large-scale review of nonlinear signal representations for EEG-based diagnosis. Benchmarked 50+ features across 7,500 signals with statistical validation. CNN-LSTM models achieving 87% accuracy on Bern–Barcelona datasets. DOI: 10.1016/j.future.2018.08.044
My passion to learn about how the world and the things around me work has been present since childhood — I've always been fascinated by machines and how they work. This has driven me to choose robotics and artificial intelligence as my fields of study. Ever since, I've been enjoying making things come to life.
Background
I'm currently a Master's student at University of Illinois Urbana-Champaign studying Autonomy and Robotics. I hold a BEng from SUTD (Singapore) with a Robotics Focus and CS Minor under the Honours and Research Programme (SHARP) with Global Distinguished Scholarship. I aim to apply my knowledge to improve lives through technology, automation, and AI.
Experience Across Environments
Research Institute — A*STAR Institute for Infocomm Research (I2R)
Multi-National Corporation — Hyundai Motor Group Innovation Centre, Robotics Center
Working with Agile/Scrum methodologies
Robot Platforms I've Worked With
Unitree G1 Humanoid · Booster K1 Humanoid
Polaris GEM e4 Autonomous Vehicle
Pioneer P3-DX Mobile Robot
Custom AGVs, AUVs, UAVs, and Ground Robots
Spoken Languages
Fluent in English, Hindi, and Marathi.
Hobbies & Interests
Building robots (of course), tinkering with electronics, photography, exploring new places and cultures, martial arts, and staying curious about how everything works.
Additional Experiences
National Service · Singapore Armed Forces · Army · Infantry · 2019 - 2021
Singapore University of Technology and Design · Student Government · ROOT · Director of Student Relations · 2021 - 2022
Singapore University of Technology and Design · House Guardian · Senior House Guardian · 2022 - 2025
