About

I am a perception engineer working at the intersection of classical 3D vision and embodied AI. My current focus is SLAM, Gaussian splatting, and VLA baselines for manipulation — problems where 3D scene understanding meets policy learning on real robotic platforms. I have built scene reconstruction pipelines using COLMAP and 3DGS, deployed VR-based teleoperation on the Unitree G1 humanoid, and trained SmolVLA on contact-rich manipulation tasks. Earlier work covered LiDAR-based perception for autonomous systems — point cloud segmentation, dynamic obstacle tracking, and sensor occlusion detection. I enjoy working close to the hardware, bridging research and deployment across the full stack. Outside of work, cricket and anime keep me sane — one for the strategy, the other for the stories. Coming from the historic city of Kurnool, I bring curiosity and a bias toward building things that actually work in the real world.

Work Experience

SLAM Engineer
Merai Newage Pvt Ltd. (A Meril Group Company) | Dec 2025 – Present

My current work sits at the intersection of 3D scene reconstruction and embodied AI. I built an end-to-end outdoor reconstruction pipeline using 30 FPS monocular captures, COLMAP-based structure-from-motion, and 3D Gaussian Splatting with MCMC-based densification — producing scene substrates used for downstream model behaviour analysis. On the robot side, I deployed a VR-based teleoperation pipeline on the Unitree G1 humanoid using Unitree’s xr-teleop package, working through dependency conflicts and network latency issues to achieve stable real-time control. I also trained a SmolVLA baseline on the LeHome dataset for cloth folding, reaching 60% / 28% task-success on train / held-out test, and built EDA pipelines and a public Gradio tool for episode inspection and joint-range analysis. Additionally, I adapted the AWS RoboMaker Hospital Gazebo world into a multi-floor simulation testbed by converting static elevator assets into dynamic, ROS-controllable models and integrating a mobile robot platform.

Software Engineer
Proliant Infotech | Sep 2024 – Jul 2025

In my role, I focused on building intelligent perception modules for autonomous systems using ROS and LiDAR data. On a daily basis, I worked on developing ROS nodes for point cloud segmentation, temporal tracking, and dynamic obstacle state estimation—achieving reliable detection of people and vehicles up to 35 meters. I also optimized data handling by converting Ouster sensor range images into compact point clouds, reducing storage needs by 80%. To improve system safety, I designed occlusion detection nodes that flagged partial or full sensor blockages in real time. Alongside this, I trained and deployed an optical flow–based ego-vehicle velocity predictor, further strengthening the overall perception stack. These contributions combined research-driven methods with practical engineering to deliver robust, real-world performance.

Computer Vision Intern
Proliant Infotech | Dec 2023 - Sep 2024

In this role, my day-to-day work revolved around developing and deploying advanced computer vision and reinforcement learning solutions. I worked with a mix of proprietary, open-source, and simulator datasets to design robust vision algorithms and integrated spatial-temporal relationships into existing models, optimizing data input and cutting training time by 50%. I also gained hands-on deployment experience by porting models to the NVIDIA Jetson Xavier NX, ensuring efficient performance on embedded hardware. Additionally, I implemented DQN-based reinforcement learning models in the CARLA simulator, exploring both urban and off-road driving scenarios to advance autonomous decision-making. This combination of research, optimization, and real-world deployment shaped a strong end-to-end workflow in AI-driven autonomy.

Education