Sakib Chowdhury

Sakib Chowdhury

I am an M.Sc. student at Stevens Institute of Technology. My research is focused on studying robot motion control in complex dynamic environments. I am focused on developing high speed controls and motion planners for robotic manipulators. I have also worked as a Machine Learning Engineer at Celloscope since December 2021. I have completed my bachelors at the Department of Electrical and Electronic Engineering in Bangladesh University of Engineering and Technology (BUET). I have also worked as an undergraduate researcher at Department of Computer Science, BUET.
My detailed resume can be found HERE.

Recents

How fast can we plan it?

Motion Planning · PyTorch · Real-time Systems

This project builds a low-latency neural motion planner for a mobile manipulator (Husky base + Franka Panda with a table-tennis paddle) that can intercept fast, moving targets—like a ball—where classical planners often miss due to planning delays. Its core value is speed and consistency: inference in ~5 ms (vs ~100 ms typical for RRTConnect) preserves crucial reaction time for tight-deadline tasks and enables stable receding-horizon replanning as targets or obstacles move. Architecturally, the planner ingests a compact 5×10 “context grid” (start/goal joints plus up to three obstacles encoded as position/orientation/size) and a 64×64 occupancy map, processes them with a dual-stream CNN (context stem + residual blocks; map CNN with pooling and adaptive averaging), fuses the features through a shared MLP, and outputs (1) a feasibility probability, (2) per-joint time breakpoints, and (3) per-joint positions; trajectories are reconstructed from ≤K_MAX RDP-compressed keypoints and interpolated for smooth, time-parameterized execution, yielding scene-independent, constant-time planning that is both fast and robust.

Model effectiveness overview

Learning to Strike Robotic Table Tennis with Residual Learning

Robotics · Control · High-speed Decision Making

Motivated by recent advances in robotic systems’ ability to interact with dynamic environments, we study autonomous robotic table tennis in this paper. We design a high speed robotic arm to play table tennis, and develop a software system that predicts the incoming ball’s trajectory and learns to control the robot arm for striking the ball to the opponent side of the table. Utilizing the idea of residual physics, we develop a residual predictor to predict a striking position for the incoming ball using sparse observation of the ball’s starting trajectory. We control the arm to the contact-ready position, and then use a neural network to output the striking velocity to hit the ball. The neural network is trained through experience using a regression-based learning method. Simulation results in a robotic simulator demonstrate the superior performance, benefiting from efficient learning

Updates

  • 12/2023 ---- "A Simulated Intelligent Pixelated Electrode Array for Surface Electromyography Sensors" Published at IEEE Sensors Journal
  • 12/2023 ---- "SynthNID: Synthetic Data to Improve End-to-end Bangla Document Key Information Extraction" Published at EMNLP Bangla Language Processing Workshop
  • 09/2023 ---- Received Provost's Doctoral Fellowship award
  • 09/2023 ---- Joined Stevens Institute of Technology as a PhD student in robotics
  • 08/2022 ---- "SpectroCardioNet: An Attention Based Deep Learning Network Using Triple-Spectrograms of PCG Signal for Heart Valve Disease Detection" Published at IEEE Sensors Journal
  • 06/2022 ---- "SHONGLAP: A Large Bengali Open-Domain Dialogue Corpus" published at LREC
  • 05/2022 ---- Promoted to full time research engineer at Celloscope
  • 12/2021 ---- Joined Celloscope as a part time research engineer


Work
Experiences

  • 2023-Present
  • Stevens Institute of Technology
  • Graduate Researcher
    • Developed and trained novel convolutional motion planners using PyTorch for ultra high-speed motion planning in mobile manipulators. Designed a simulation training environment in PyBullet (Python wrapper of the Bullet physics engine) for a mobile manipulator system: a Franka Emika Panda arm mounted on a Husky robot. Trained the motion planners in simulation and successfully transferred them to the real Franka Emika Panda robotic arm for deployment. Applied ONNX quantization to optimize the neural motion planning model for faster inference on NVIDIA A6000 GPUs, achieving 10x speedups over classical planners (e.g., RRT, PRM) in the ROS MoveIt framework. I have also studied motion planning and control of robotic manipulators in complex dynamic environments. My research was focused on bridging the gap between ideal and realistic physics with the help of learning based systems. This work was partially supported by the US National Science Foundation.
  • 2021-2023
  • Celloscope
  • Machine Learning Engineer (AI)
    • We have developed the first voice banking system in Bangladesh here at Celloscope. It is a voice-enabled system that allows users to do regular banking tasks (such as balance transfer and balance inquiry) just by using their voice. The responses are also generated as a synthetic voice in order to provide a voice-to-voice conversation experience. This system is currently integrated into the voice banking app of Agrani Bank. We have also developed a production-grade license plate detection and recognition system at Celloscope which is deployed at a toll plaza in Bangladesh. Our developed system uses document understanding transformers instead of traditional convolutional OCR systems. Document Understanding Transformers are multimodal systems that allow querying on the information in a jpeg file.
  • 2021-2022
  • Department of CSE, BUET
  • Research Assistant (Part-Time)
    • We have developed a system that detects derailment from upto 1200 meters distance by sensing the vibrations generated from the movement of train. Dr. A. B. M. Alim Al Islam, Professor, Department of CSE, Bangladesh University of Engineering and Technology (BUET) is the supervisor of this project.
  • 2020-2021
  • Adorsho Pranisheba
  • Intern Engineer (IoT)
    • During my internship period, I was involved in the development of BOLUS - a cattle health monitoring tool for milk farms. BOLUS is a small piece of hardware that is placed inside the stomach of domestic cows and it stays there for 5 years. During its lifespan, it sends movement data, temperature information, and other health-related data of cows to a remote server where an AI system identifies the current health condition of the herd.

Education

Professional
Affiliations

  • 2017-Present
  • IEEE
  • Member

Award &
Achievements

  • Industrial Automation Challenge 2017 (BUET Robotics Society)
  • Champion
  • The problem statement for the competition was to design a robotic arm with visual capability to identify the shape of objects coming through a convayer belt. Moreover, the arm had to sort the objects based on their shapes. Our team outperformed all the other teams by successfully completing without any error.