cv

Masters of Science in Computer Science (CGPA - 4.0 / 4.0)

University of Colorado Boulder

Bachelors of Technology in Electronics and Communication Engineering

Indian Institute of Technology (IIT) Guwahati, India

RMap Millimiter-Wave Radar Mapping through Volumetric Upsampling (under review at ICRA 2024)

Tell Me Where to Go A Composable Framework for Context-Aware Embodied Robot Navigation (Accepted at CoRL 2023)

MultiNavCon An Architecture for Language-Directed Navigation of Multiple Robots (under review at ICRA 2024)

Grouped BERT for Multi-Label classification to reason the human values behind the arguments

Research Assistant

Autonomous Robotics and Perception Group

• Developed the state-of-the-art generative transformer, UpPoinTr, for enhancing volumetric maps from sparse and noisy radar scans, surpassing the prior models by 8% in performance and generate lidar-like navigable maps
• Designed NavCon, a low bandwidth robot navigation framework using Large Language Models(LLMs), achieving a 71.3% success rate in guiding a Spot robot through intricate human commands in various environments
• Utilized only the radar data, sparse pointclouds compared to Lidar data, to overcome the challenges in visually degraded scenarios and improved the odometry estimation by 8% , using transformer and DeepVO architectures

Research Assistant

Vignesh Kasinath Lab

• Devised Multi-UNet architecture, seamlessly integrating the segmentation of multiple structures, resulting in a substantial 13% boost in F1-score
• Designed U-NeXt architectures, combining the ConvNeXt and U-Net, specifically tailored for tomograms captured at different scales, resulting in a f1 score of 85% for segmentation

Computer Vision Research Engineer

Samsung Semiconductord India R&D

• Developed 3D scene reconstruction algorithm, exclusively leveraging depth from ToF sensors, achieving a real-time processing speed of 20fps
• Improved the accuracy by 5% by detecting and removing outliers and also automated the pose alignment of output 3D scene with groundtruth to eliminate the manual alignment in MeshLab
• Reduced the latency of Remosaic deep learning models for 200M pixel camera sensor using quantization and pruning techniques by 10 % with an unnoticeable degradation of 0.1% in perceptual quality
• Optimized Samsung’s CMOS camera sensor noise reduction algorithm with OpenMP parallel programming, thereby reducing the algorithm runtime by 33%

Software Engineer

Qualcomm

• Spearheaded the development and maintenance of python audio library to evaluate both the objective and perceptual audio quality of the Bluetooth headsets
• Enhanced python automated test framework with new features that populate test vectors and visualize audio output signals, leading to a 10%-15% reduction in both the validation and development teams’ efforts

Research Intern

Eagle Eye Networks

• Leveraged YOLO for object localization, optical flow for trajectory estimation, achieving 83% accuracy in detecting location and trajectory anomalies through Incremental Spherical Clustering and frequency heatmap

• Received Employer of the Month at Samsung, for successfully optimizing the noise reduction algorithm to reduce runtime by 33%

• Received the Spot Appreciation award, for exceptional performance in delivering the 3D reconstruction pipeline for Samsung's ToF sensors

• Received Qualstar from Qualcomm for continued excellence in improving the audio quality analysis framework