Anik Sarker
Hello! I'm currently in my third year pursuing a PhD in Mechanical Engineering at Virginia Tech. I've been fortunate to work at the Assistive Robotics Laboratory, under the supervision of Dr. Alan Asbeck since Fall 2021. During my Masters, I had the privilege of being supervised by Dr. Nilanjan Chakraborty at the Interacting Robotic Systems Laboratory. I'm deeply interested in exploring the realms of Robotics and AI.
I've a wide range of research interests. On a higher level, my research revolves around the domain of Robotics, Computer Vision, and both Deep Learning and Machine Learning. Additionally, my background as a mechanical design engineer and as an educator has given me a unique perspective, making it easier for me to grasp new concepts. From a broader perspective, I am passionate about enabling robots to perceive the world more accurately and make intelligent decisions to interact more effectively with humans.
(1) Point Cloud Registration Algorithm:
Developing novel algorithms for correspondence-free point cloud registration . We are rigorously testing our algorithms against benchmark methods like TEASER, ICP and Deep Learning based methods such as PointNetLK, Deep Closest Point, etc. Additionally, we're expanding our work to include spherical image registration, leveraging our fast spherical correlation algorithm (detailed in the paper section). Our exploration also extends to related applications such as pose estimation, 3D reconstruction, and image processing.
This research is funded by the NSF.
(2) Hand / Finger pose estimation using IMU:
Our project involves deriving ground truth data from dual-camera reconstructions of hand poses. We focus on extracting precise information about the forefinger and thumb pose from this data. Our approach includes training deep learning models, specifically Transformers, to accurately map temporal data from IMU sensors to predict the finger pose over time. Key aspects of our work encompass camera-sensor synchronization, calibration, data processing, basic digital signal processing, camera-based hand pose reconstruction, and the application of deep learning techniques.
The project is being funded by NSF.
(3) Human Upper-Body Pose Estimation Using IMU and RSSI:
This is an extensition of the previous work. Prevoiously we used 3 IMUs to estimate kinematics of human upper-body. In this work, we are experimenting with both IMU and RSSI (signal transmitters and receivers) to evaluate the feasibility of human pose estimation.
The project is being funded by NSF.
Fast Spherical Correlation and Probabilistic Feature Matching for Orientation Alignment
Anik Sarker, Alan T. Asbeck
2024(Under Review)
Webpage (coming soon) •
PDF (coming soon) •
Fast spherical cross-correlation technique offering O(n) complexity, ensuring robust vector alignment even in the presence of outliers.
Novel Alogirthm, Transformed Basis Vector Alignment via Probabilistic Correlation (TBVA-PC) is developed for 'Automatic Calibration' of multi IMU systems.
Capturing Upper Body Kinematics and Localization with Low-Cost Sensors for Rehabilitation Applications.
Anik Sarker, Don-Roberts Emenonye, Aisling Kelliher, Thanassis Rikakis, R. Michael Buehrer, Alan T. Asbeck
Sensors 22(6): 2300 (2022)
Webpage •
PDF •
On Screw Linear Interpolation for Point-to-Point Path Planning.
Anik Sarker, Anirban Sinha, and Nilanjan Chakraborty
IEEE International Conference on Intelligent Robots and Systems (IROS 2020)
Webpage •
PDF •
Task Space Planning With Complementarity Constraint-Based Obstacle Avoidance.
Anirban Sinha, Anik Sarker and Nilanjan Chakraborty
ASME 2021 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (IDETC-CIE 2021)
Webpage •
PDF •
RIS-Aided Kinematic Analysis for Remote Rehabilitation.
Don-Roberts Emenonye, Anik Sarker, Alan T Asbeck, Harpreet S Dhillon, R Michael Buehrer
IEEE Sensors / August 2023
PDF
Novel method of orientation estimation using Reconfigurable Intelligent Surfaces (RISs) for enhanced stroke rehabilitation.
Advanced position and orientation tracking with superior near-field accuracy, leveraging existing wireless signals for critical in-home health data collection.
Geometric motion planning in task space using complementarity constraints to avoid collisions.
Anik Sarker
Stony Brook University, December 2018
Webpage •
PDF (can be provided upon request);
Object Detection: Fine-tune custom dataset with YOLO V8, NAS, Facebook Detr
The goal of this project was to explore the performance of different object detection and tracking algorithms on a custom dataset. Fine-tuned "SODA10M" datset (the largest 2D object detection dataset in autonomous driving) using state of the art "YOLO-NAS", "YOLO-V8" and "Facebook Detr".
Applied sevelral unique vizualization techniques to compare performances of differnet modes both quantatively and subjectively.
Human pose reconstruction and ergonomic pose classification:
Human pose reconstruction and ergonomics evaluation have many other applications, such as stroke rehabilitation, underground mining, safety and rescue working, bio-mechanical monitoring, sports and fitnesstraining, assistive devices, and virtual reality interaction.
In this project, we present an approach for Human Pose Reconstruction and Human Posture (Ergonomics) Evaluation based on five low-cost IMU sensors (XSens DOT). This project has two main directions. Firstly, Pose Reconstruction: we use information from 5 IMU sensors (orientation, acceleration) to predict full-body kinematics (23 human body segments). Secondly, Ergonomics Evaluation: Based on the reconstructed pose, we demonstrate a computational framework to classify two postures, namely walking and sitting postures into two categories (normal or poor posture). We train two auto-encoders (sequence-to-sequence, transformers) to map the latent space (five segments) information to the full pose-space (23 segments). For ergonomics evaluation we consider 3 ergonomics parameters, which are (1) Flexion (2) Lateral Bending (3) Extension for ‘Pelvis’ and ‘T8’ segments.
For both pose reconstruction and ergonomics evaluation, we compare reconstruction error and classification error with respect to the ground truth.
Sequence-to-sequence, Transformers, PyTorch, XSens (PDF)
B-SPLINE Global Surface Interpolation and Approximation:
Developed a GUI app for demonstrating both B-Spline global surface interpolation and approximation.
The user interface allows user to input thogh mouse and keybord to control several parameter related to surface interpolation and approximation.
C++, OpenGL (PDF)
2023
On November 14th, passed PhD preliminary exam. I am a PhD candidate now!