Ajay Narasimha Mopidevi
CS Grad Student at University of Colorado Boulder | Computer Vision Engineer at Samsung
I am a highly skilled professional with over 5 years of industry experience, specializing in Deep Learning for Robotics Perception. I have a strong academic background, demonstrated by an exceptional 4.0 GPA and other academia projects, and a proven research background with 2-lead authored publications and 1 co-authored publication.
I am currently working on radar-based navigation and mapping to enable the perception systems to function properly even in extreme weather conditions like smoke, fog etc. Even though radar sensors are proven to be effective in these environments, they pose additional challenges like sparse data representations and multi-path reflections. I have developed the state-of-the-art generative transformer, UpPoinTr, for enhancing volumetric radar maps surpassing the prior models by 8% in performance and generate lidar-like navigable maps. My current contributions have been submitted to IROS 2024. and are currently under review. I am also working on integrating natural language instructions with landmark-guided navigation to enhance the search and rescue operational robots. We used vast prior information in Large Language Models to model these instructions and guide the robots to the desired location. Our work has been accepted to the CoRL 2023 conference.
I am also working with Kasinath’s lab at CU Boulder to segment the cellular structures like ribosomes, membrane, micro-tubules, filaments from the Cryo-ET tomogram. Cryo Electron Tomography combines cryogenic sample preparation, electron microscopy and computational image analysis to visualize the internal cellular structures and macro-molecules in their native state. Limiting the annotaions to <1% of entire images, we used machine learning techniques like Auto-encoders and U-Net to segment these structures.
During my time at Samsung, I developed computer vision algorithms, combining traditional and deep-learning methods, to enhance the performance and accuracy of Samsung Camera sensors. A significant accomplishment was my work on real-time 3D scene reconstruction algorithms utilizing only the range data from ToF sensors. I have consistently delivered impressive results and further expanding my knowledge in the domain by working on diverse projects. For my remarkable contributions to various projects, I have received various accolades, including Employee of the Month.