About Me
I am a Masters Robotics graduate, with focus on Computer Vision from Northeastern University with over three years of experience in Machine Learning and Computer Vision.
My expertise lies in building end-to-end pipelines using advanced 3D vision techniques like SLAM, Structure from Motion, Pose Estimation, and Depth Estimation, with research and industry experience.
I am passionate about leveraging 3D vision and deep learning to develop innovative solutions that bridge the gap between research and real-world applications.
Publications
Logarithmic Lenses: Exploring Log RGB Data for Image Classification
B. Maxwell, A. Patel
Computer Vision and Pattern Recognition Conference (CVPR), 2024
View PublicationProfessional Experience
Systems Vision Engineer
Medtronic
Sept 2025 – Present
Boston, MA
- •Designed a Pose Graph pipeline in GTSAM using Robot Kinematics and TensorRT based YOLOv8 vision keypoints to reduce the drift and track the instrument points accurately in Out-of-View scenarios
- •Engineered a Arm Collision Avoidance System using Trimesh of the medical robot using URDF and Kinematics using RTI DDS publish-subcribe method for real time visualization in Open3D
- •Established a secure bridge communication between Production and Engineering Robotic systems to effectively transfer Kinematics messages
Research Assistant - AirLab
Carnegie Mellon University
July 2025 – Nov 2025
Remote
- •Integrated Relative Pose Graph Optimization in ROS2 in C++ using GTSAM Fixed-Lag Smoother in the IMU Preintegration module of the multi-modal IMU-LiDAR sensor fusion to reduce long term drift in SLAM
- •Achieved 35.8% lower ATE and 52.5% RPE on the SubT-MRS Laurel Cavern dataset with Velodyne LiDAR
- •Executed trajectory mapping using Livox LiDAR and IMU sensors on the Unitree G1 robot, applying a low-pass filter to mitigate IMU bias and enhance mapping accuracy
- •Built a Thermal-Intertial Odometry system with uncertainty aware-feature weighting for pose estimation in CART dataset
- •Enhanced an Uncertainty SLAM system by integrating Voxel Maps, resulting in improved localization robustness and accuracy in challenging environments
Surgical R&T Machine Learning Engineer
Medtronic
Jan 2024 – Apr 2025
Boston, MA
- •Built an end-to-end SLAM pipeline with DROID-SLAM for dense depth estimation in surgical videos, optimizing camera trajectory using GTSAM and refining 3D reconstruction with Bundle Adjustment and LightGlue feature matching
- •Developed a real-time Ground Truth pose estimation pipeline using OptiTrack camera capture and robot kinematics with PnP and ROMA feature detection for training deep learning models on instrument articulation
- •Performed Semantic Segmentation for, Robot-Assisted Surgery, on 10,000 medical images from S3 bucket, containing both mask and line annotations, to segment hernia using the Swin Base Transformer
- •Developed a YOLOv8-based pipeline for precise detection of surgical instrument tips from medical images in real-time
- •Applied Monocular Depth Estimation to get metric distance between two instruments from an image by Depth Anything
- •Implemented a PyTorch wrapper with Optical Flow on FAST API using Unimatch, converting models to ONNX and TensorRT for 10x reduction in real time annotation of medical image frames with 1-second latency
Research Student - Computer Vision
Northeastern University
May 2023 – Dec 2023
Boston, MA
- •Researched raw log RGB data's impact on deep networks like ResNet-18, improving classification performance and robustness to intensity and color variations
- •Codeveloped the novel RAW10 dataset (10k DNG & JPG images each, 10 categories) to advance LOG RGB research in Computer Vision community
- •Published CVPR 2024 paper on this research
Artificial Intelligence Engineer
Kisan Drip Irrigation Pvt Ltd
Aug 2020 – Aug 2022
India
- •Integrated ElasticFusion: RGB-D SLAM with C++ to align multi-view point clouds from Intel RealSense D455 cameras,enabling accurate 3D Reconstruction for pipe inspection and defect analysis
- •Experimented with PointNet-based deep learning models in Python for point cloud classification to enhance complex defect identification, achieving a 30% improvement over traditional 2D vision methods
- •Implemented YOLO-based object detection to localize drippers in pipe assemblies, enabling precise hole punching
- •Deployed the 3D vision pipeline as a containerized FastAPI service integrated into on-premises manufacturing workflows
Education
Northeastern University
Master of Science in Robotics ( specialization in Computer Vision)
GPA: 3.6/4.0
September 2022 – December 2024
Boston, MA
Relevant Coursework:
Robotics Sensing & Navigation, Advanced Computer Vision, Autonomous Field Robotics
Featured Projects
Technical Skills
Programming
PythonC++CUDA
Frameworks & Libraries
PyTorchOpenCVOpen3DVTKROS2GTSAMG2OOptiTrack SDK
Computer Vision & Deep Learning
Semantic SegmentationObject DetectionMonocular Depth EstimationPose Estimation3D ReconstructionSLAM Bundle Adjustment
Simulation & Visualization
CARLAGazeboRVizThree.js
Cloud & Deployment
AWS S3ONNX RuntimeNVIDIA TensorRTPyTorch DDPFastAPI