Modabbir Adeeb
Robotics Engineer | MEng in Robotics
madeeb@umd.eduAbout
I’m a robotics engineer with a background in mechanical engineering, currently pursuing a Master’s in Robotics at University of Maryland. I work on machine learning, computer vision, and autonomous systems, with projects in 3D reconstruction, path planning, and deep reinforcement learning. I enjoy solving real-world problems through software and hands-on experimentation.
Contact
- email madeeb@umd.edu
- linkedin linkedin.com/in/modabbiradeeb2001
- github github.com/modabbir22
- grabCAD grabcad.com/modabbir.adeeb-3
Work Experience
- Jun 2024 - Aug 2024
Reinforcement Learning Intern
Sommer AIJun 2024 - Aug 2024
Developed and optimized a two-stage rebar detection algorithm using YOLOv5-P6m deep learning and DBSCAN clustering to enhance detection accuracy. Achieved 92.4% precision and 97.4% recall for rebar intersection detection, significantly improving identification stability by 32.5%
- Jun 2021 - Aug 2021
Technical Engineer Intern
Mitsubishi Electric Elevators and EscalatorsJun 2021 - Aug 2021
Assisted the maintenance team in testing the condition of the control systems with oscilloscopes and testing emergency operations using MelEye software during quarterly visits. Evaluated the performance efficiency of the predictive maintenance system by applying alternative algorithms to the pre-trained model – attained the highest accuracy of 95.8%.
Projects
-
Multi-view 3D Reconstruction using Transformers
OpenCV
Nov 2024 – Dec 2024
- Developed a Transformer-based 3D reconstruction model trained on ShapeNet, processing single-view 2D images using a pretrained Data-efficient Image Transformer (DeiT) encoder and a Transformer Decoder to generate 3D voxel representations.
- Enhanced accuracy and efficiency by integrating a CNN Decoder and optimizing performance using Dice Loss.
- Achieved an IoU score of 0.439, performing close to SOTA models with improved efficiency.
-
Motion Control and Path Planning for Mobile Robots
ROS2
Jan 2024 – May 2024
- Developed a planner package enabling a Turtlebot to navigate mazes using 8 algorithms: DFS, BFS, Dijkstra, A*, RRT, RRT*, Informed-RRT*, and Bezier Informed-RRT*.
- Compared algorithm performance and achieved autonomous navigation in custom Gazebo environments.
-
Ground Detection using Semantic Segmentation
OpenCV
March 2024 – May 2024
- Trained a U-Net model to segment drivable paths using video data across various scenarios (pedestrian paths, highways, residential streets).
- Achieved 78% segmentation efficiency.
-
Autonomous Robot Navigation using Deep Reinforcement Learning
ROS2, Docker
Feb 2024 – April 2024
- Implemented goal-based Turtlebot navigation in unknown environments using the twin delayed DDPG algorithm with Velodyne LiDAR.