Enhanced monocular metric depth estimation by learning an image-conditioned scaling factor that corrects global scale bias, enabling accurate 3D perception without relying on additional sensors.
Developed and compared MDP and POMDP-based reinforcement learning agents to navigate an autonomous vehicle through an uncontrolled intersection with other agents of unknown, latent driving styles.
Trained a robotic manipulator for pick-and-place tasks using an action affordance model, achieving high success rates on both seen and unseen objects.
Configured Unitree Go1 with 3D LiDAR and onboard Direct LiDAR-Inertial Odometry and Mapping (DLIOM) to enable accurate state estimation and autonomous navigation using ROS2 and Nav2 in real-world environments.
Controlled a differential wheeled robot with a joystick by establishing bluetooth connections with the robot and joystick and mapping the joystick state to robot wheel speeds.
Created a program that converted input images (uploaded directly by users or captured with a robot’s fisheye camera) to trajectories that the robot would follow to perform its drawing task.
Trained and tested deep learning based object detection models, RetinaNet and YOLOv5, on an on-road vehicle dataset and their performance was investigated.
Designed an underwater localization system for UUVs utilizing inertial pose estimation and beacon triangulation based on acoustic communication understanding constraints of underwater communication and dead reckoning using IMU.
Created a map of school tunnel using various SLAM algorithms on sensor data collected by a LiDAR, Cameras, and IMU.
Utilized Gmapping SLAM, Frontier-Based Exploration and tag detection on TurtleBot3 to achieve a reconnaissance mission autonomously in a closed and unkown environment.
Designed a wearable fitness tracker using an IMU sensor and complementary filter to estimate body orientation and motion in real time. Implemented posture classification and repetition counting algorithms for exercises (e.g., squats), using calibrated angle and angular velocity thresholds to detect form and timing accuracy.
Designed a flower-care monitoring system using various sensors such as moisture, light, and temperature sensors and transmitted sensor data via Wi-Fi to a Google Cloud server for further analysis.
Configured a robotic arm on an FPGA development board using memory-mapped I/O. Programmed control logic in C++ to interpret directional gestures from a Wii-mote’s onboard IMU. Detected up, down, left, and right motions to drive the arm’s servos in real time. Mapped Wii-mote button inputs to control the gripper’s open and close actions. Integrated sensor data processing, embedded control, and wireless communication for gesture-based robotic manipulation.