Publications

TL;DR: FALCON enables various heavy-duty humanoid loco-manipulation tasks via a new dual-agent force-adaptive RL framework.

Projects & Research

Built an autonomy system that fuses 6D pose estimation (NVIDIA FoundationPose) with motion-capture localization for precise perception and navigation. This allowed the Unitree G1 to autonomously manipulate totes and operate effectively in real-world factory workflows.

Built and deployed Diffusion Policy leveraging RGB-D and state inputs for robust real-world block pushing on a Franka Panda robot. Enhanced training through human-in-the-loop demonstrations collected via a MoveIt model-based controller.

Developed perceptive locomotion on Unitree Go1 with a real-time mapping pipeline (RTAB-Map + robot-centric elevation mapping) on Jetson NX for 20 Hz terrain awareness. Combined it with NMPC (OCS2) and whole-body control to achieve reliable traversal on challenging, uneven surfaces.

Designed a compliant robot torso by 3D-printing TPU intervertebral discs that provide passive damping and safety. Built a parallel-manipulator actuation system delivering 3-DoF rotational motion with compact packaging and high rigidity under load.

Developed a portable single-channel EEG device with custom PCB design (BLE SoC) and 3D-printed enclosure, enabling audio closed-loop stimulation during sleep. The system was used to monitor overnight neural signals and collect high-quality datasets. These recordings supported the development and training of a sleep-stage classifier.

Proposed an efficient deep learning model and a novel technique to effectively extract time‐invariant features for sequence residual learning in down‐streaming modules based on raw single-channel EEG.

Programmed the ABB IRB 910 with MATLAB Robotics Toolbox to implement gravity compensation and torque-based control, improving pick-and-place precision by 25%. Developed dynamic models using Newton-Euler and Lagrangian formulations in MATLAB to enable accurate motion control and reduced trajectory errors.

Developed a TurtleBot platform capable of long-distance SLAM with 2 cm drift using G-mapping and AMCL in ROS. Added A* and RRT-based planners for optimal exploration and obstacle avoidance, and enhanced robustness with Kalman Filter re-localization after external disturbances.

Implemented forward and inverse kinematics with cubic/quintic trajectories to control a 3-DoF robot manipulator for precision object sorting. Enhanced perception by integrating YOLOv5 and HSV thresholding in OpenCV, achieving >97% detection accuracy. Built a URDF and synchronized real-time CAD animation with the robot’s motion.

Education

cloned from here!