MEng Robotics at UIUC, building learning-based systems for humanoids, autonomous vehicles, and mobile robots.
Terrain-aware locomotion for a 29-DOF Unitree G1, trained with PPO in Isaac Lab under a stand→walk→DR curriculum. A CNN encodes height scans alongside proprioception into the actor-critic, with domain randomization across mass, friction, PD gains, and pushes.
Real-world deployments, novel algorithms, and sim-to-real systems across humanoids, AVs, and mobile robots. Featured projects showcase the deepest work.
Conversational navigation on a Unitree Go1 + Jetson Orin: a fine-tuned NaVILA (8B VLA) emits mid-level language actions (forward / turn / stop) from RGB streams. Added a new "ask user" action token so the policy defers under ambiguous instructions, asks a GPT-4o-generated clarification question, and resumes after the operator answers via a dual-variant instruction amender. Diagnosed that NaVILA expects an 8-frame linspace memory bank rather than a sliding window - correcting it raised offline accuracy ~51 points. LoRA fine-tuned over teleop + 4 DAgger iterations with ~1 Hz offboard inference and 8-bit quantization on an RTX workstation; the ask-user checkpoint is the first to fire the action correctly on the live robot.
A conversational extension of NaVILA on a Unitree Go1 quadruped: the vision-language-action policy defers under ambiguous instructions, asks the operator a clarifying question, and resumes - all in the language modality it already speaks.
Precision USB-A insertion on a UR5e: a CV port localizer + IK reaches a pre-insert pose, then a SAC residual policy layers joint-delta corrections using F/T feedback during contact. Trained in MuJoCo with domain randomization over friction, mass, port pose, and F/T sensor noise; the F/T-noise DR was the load-bearing factor for clean sim-to-real - the policy transferred zero-shot to the real UR5e and consistently completes the insertion. Also benchmarked an OpenVLA-based learned base controller against the classical CV + IK stack.
Precision USB-A insertion on a UR5e under tight clearance, learned via residual reinforcement learning layered on top of a classical CV + IK base controller.
AutoShield successor on the UIUC Polaris GEM e4. A MID-style diffusion Transformer (DDPM-trained, DDIM 10-step inference, joint multi-agent cross-attention) predicts pedestrian trajectories over the AutoShield LiDAR + RGB-D tracker, running live on-vehicle. An MPPI motion planner (K=600 rollouts, joint steering+acceleration) and a text-promptable goal-selection module (YOLO-World / LangSAM open-vocabulary detection, LiDAR-fused to an MPPI goal pose) each validated in simulation.
AutoShield successor on the UIUC Polaris GEM e4: a diffusion-based pedestrian predictor, an MPPI motion planner, and a text-promptable goal-selection module.
Open-vocabulary 6D tracking extending FoundationPose with Moondream2 VLM scene analysis, SAM-3 text-prompted segmentation, and hierarchical mesh acquisition (GT → Objaverse-XL retrieval → TripoSR generation). Composite scoring (IoU + Depth + Silhouette) selects best proxy; achieves 100% ADD-S AUC on texture-rich objects and supports language-driven dynamic target switching mid-task on YCB-Video.
Open-vocabulary 6D pose tracking driven by natural-language prompts, extending FoundationPose.
Replaced separate EfficientNet + BERT encoders with a frozen unified CLIP backbone (128-dim adapter projection) in the HumanVLA teacher-student pipeline, resolving catastrophic forgetting observed when fine-tuning visual encoders. Behavior Cloning + DAgger distillation across 615 episodes in 4 HITR room types; 62.1% success on unseen tasks (+1.9% over baseline, −4.4% placement error).
A language-guided humanoid policy that navigates HITR rooms and rearranges objects from natural-language instructions.
Modular ROS 2 autonomy stack on the UIUC Polaris GEM e4: Ouster OS1-128 LiDAR → DBSCAN clustering + EMA tracking fused with OAK-D LR RGB-D via YOLOv11 detection (weighted 0.8/0.2 distance, 0.3/0.7 bearing). Forward-simulated TTC drives a 3-state safety FSM; Stanley lateral + PID longitudinal control with PACMod2 hard-brake override. 91% field success.
Pedestrian-intent prediction and safety-filtered driving on the UIUC Polaris GEM e4. Owned the pedestrian stack in a 4-person team.
Two-tier PPO hierarchy for dual Unitree G1 humanoids in IsaacLab: low-level [256,256] MLP drives 29-DOF joint-position-delta actions through a 7-phase standing→omnidirectional curriculum (30M steps, 256 envs); high-level [64,64] navigation policy converges agents from 10 m to 0.5 m with 95% success. 25+ shaped reward terms enforce gait symmetry, contact penalties, and energy regularization.
Two coordinated Unitree G1 humanoids walking toward each other under a hierarchical PPO policy in IsaacLab.
Full-stack embodied AI on physical Booster K1 Humanoid: Gemini VLM semantic goal planning with ROS2 Nav2 autonomous navigation and Snap AR Spectacles AR collaboration.
Embodied AI on a physical Booster K1 humanoid, paired with a Snap AR Spectacles co-pilot.
Led 7-person team. Kalman-filtered admittance controller maintaining <150 N brush-to-glass contact. Full ROS2 stack on Jetson Orin Nano with SMACH FSM. Validated in live Changi Airport deployment.
AGV that climbs to 2 m and cleans curved glass at Changi Airport. SUTD capstone, 7-person team lead.
Novel PP variant incorporating bicycle-model lateral error decomposition (cross-track slippage + heading error) into the control law. DIAPP adds a dynamics correction term c_d·e_lat,total to the steering output; RDIAPP extends this with RPP-style curvature-regulated velocity. At 70 km/h on an S-track, RDIAPP achieves 1.79 m max CTE vs. PP's 30.26 m (diverges); on a U-track, 2.49 m RMS CTE vs. 31.6 m for PP.
Novel Pure Pursuit variant that doesn't diverge on consecutive high-speed curves. SUTD SHARP thesis.
Autonomous lane-following on Jetson Xavier + ZED2: dual-filter perception (Gaussian-weighted adaptive grayscale + HLS color space) with bitwise AND fusion targeting 8% pixel density, vanishing-point radial-scan histogram rejecting racetrack markings, and sliding-window 2nd-order polynomial fit. Custom proportionate lateral controller with distance-weighted steering (f(c_y) = C/(c_y − h_y + ε)). Completed full 12-min autonomous laps across variable lighting conditions.
Outdoor autonomous lane-following on a scaled 4WD race car, robust across changing light.
Autonomous unknown-maze exploration on TurtleBot3 (LattePanda): frontier-based exploration with DFS cluster grouping and centroid-based goal selection, A* costmap planner with B-spline path smoothing, Pure Pursuit trajectory follower, and fine-tuned YOLOv11n with a novel spatial deduplication tracker projecting bounding-box centroids into the map frame via LiDAR range lookup for unique instance counting.
Autonomous maze exploration with object instance counting on a TurtleBot3.
Software Lead. Cascaded PID for 6-DOF underwater stability with IMU/depth fusion via state-space control. Real-time CV for autonomous target acquisition. SAUVC 2023 Finalist.
Autonomous underwater vehicle for the SAUVC 2023 competition. Software Lead, SAUVC 2023 Finalist.
Low-cost quadcopter with off-the-shelf components. Integrated propulsion, power distribution, and flight electronics for stable teleoperation.
Personal build: a low-cost teleoperable quadcopter from off-the-shelf parts.
Custom PCB with motor control, dynamic air-cooling, slope-based actuation, and anti-roll basket mechanism for incline stability.
Self-stabilizing electric grocery trolley.
CPG-based motion control for a soft batoid robot with sinusoidal X/Pitch/Roll propulsion and strain-gauge obstacle detection.
CPG-driven motion control for a soft batoid-style underwater robot.
Low-cost EM leak detector with Bluetooth connectivity for non-invasive monitoring of variable-length fluid pipes.
Low-cost electromagnetic leak detector for variable-length fluid pipes.
Angsana seed-inspired autorotating craft with single-flap motor trajectory control and aerodynamic calibration for controlled descent.
Autorotating aerial craft inspired by the Angsana seed.
8-bar lifting mechanism with novel intake for cube-frame collection; Best Design Award at Asia-Pacific Robotics Championship.
Competition robot for cube-frame collection.
Benchmarked 7 monocular/stereo VO algorithms (DSO, SVO, CNN-SVO, DF-VO, TartanVO, ORB-SLAM3, DROID-SLAM) on Oxford, Munich, and Singapore rain datasets via ATE. Our DROID-SLAM + CGRP + Heuristic variant achieved the lowest stereo long-range ATE; DF-VO best monocular under 500 m. Published at IEEE CASE 2023.
VO benchmarking under rain across 7 algorithms and 3 cities.
Benchmarked 9 nonlinear feature families (MMSE, RQA, DFA, entropies, FD, Hjorth, Hurst, LLE, LZC) across 7,500 focal/non-focal EEG signals. LS-SVM (polynomial-3, 10-fold CV) achieved 87.93% accuracy / 89.97% sensitivity. Published in FGCS, Elsevier 2019.
Nonlinear feature benchmarking for focal vs. non-focal EEG classification on the Bern-Barcelona database.
CNN-RNN time-series model predicting Cat 1 Lightning Risk from weather-station features: rainfall, wind speed, temperature, humidity, and wind direction.
Time-series forecasting of Category 1 Lightning Risk warnings from weather-station signals.
BiLSTM + DistilBERT classifier for IMDA movie-review sentiment, combining contextual transformer embeddings with sequential modeling. ~96% accuracy beating classical NLP baselines.
Sentiment classifier on IMDA movie reviews.
HMMs with first- and second-order Viterbi decoding for entity recognition and sentiment tagging in informal news text.
Sequence labelling on informal news text with Hidden Markov Models.
Production deployments, real-vehicle autonomy, and simulation at scale - from 200+ AMR fleets to port-side AV trials.
All content respects applicable NDAs and Codes of Conduct - no confidential information or internal source code developed during these engagements is shown.
Full-stack robotics - from perception and planning to learning and deployment.
A focused academic path through engineering, robotics, and CS.
Peer-reviewed contributions to robotics and biomedical AI.
Comprehensive evaluation of 7 VO algorithms (DSO, SVO, CNN-SVO, DF-VO, TartanVO, ORB-SLAM3, DROID-SLAM) on monocular and stereo setups across Oxford RobotCar, 4Seasons (Munich), and Singapore heavy-rain datasets. Proposed DROID-SLAM + CGRP + Heuristic variant achieving lowest stereo ATE for long-range rain localization; DF-VO identified as best monocular approach for <500 m.
Download PaperFirst systematic comparison of 9 nonlinear feature families (52 features, all p < 0.01) for focal vs. non-focal EEG classification on the full 7,500-signal Bern-Barcelona database. LS-SVM (polynomial-3, 10-fold CV) achieved 87.93% accuracy / 89.97% sensitivity. MMSE identified as top-ranked discriminator; proposed recurrence, bispectrum, and cumulant plots for visual class separation. DOI: 10.1016/j.future.2018.08.044
Download Paper
What I really care about is humanoid locomotion. Most of my time goes into a terrain-aware RL policy I'm training on a Unitree G1 in Isaac Lab (PPO, domain randomization, and a stand-to-walk curriculum across 15 procedural terrains), with the goal of eventually transferring it to real hardware. Around that, I work on multi-agent humanoid coordination, VLA navigation and manipulation, and full-stack AV autonomy.
I come at this from a deployment-heavy background, having shipped a 200+ AMR production fleet at Hyundai, full autonomy on a Polaris GEM e4, and a cleaning robot that's live at Changi Airport. Real hardware is where I want this work to land.
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. My industry experience spans production fleet deployment of 200+ AMRs at Hyundai, AV simulation at Venti Technologies, and perception research at A*STAR. I'm driven by shipping systems that work in the real world.
Fluent in English, Hindi, and Marathi.
Building robots (of course), tinkering with electronics, photography, exploring new places and cultures, martial arts, and staying curious about how everything works.
