CV

Undergraduate Researcher

Robotics Research Center (RRC), IIIT Hyderabad

Advisor: Prof. K. Madhava Krishna. Collaborator: Prof. Sourav Garg. Robotic perception, 3D vision, and planning & navigation.

• Plann3r — 3D-grounded planning costs (CoRL 2026, first author). Fine-tuned VGGT with a new prediction head that outputs goal-conditioned geodesic costmaps for navigation from single-camera RGB input; evaluated on HM3D scenes.
• Designed and ran the Plann3r rebuttal experiments, including a loop-closure extension addressing single-traversal assumptions, retrieval-noise sweeps, and bootstrap confidence intervals.z
• PixelLoop — pixel-level loop closures (IROS 2026). Co-developed a loop-detection pipeline that adds pixel-level loop closures to topological navigation methods such as MASt3R-Nav, and ran its benchmark evaluation.
• Ongoing. Topological navigation, visual place recognition (VPR), and RGB-only last-mile docking for mobile robots.
• Deep-learning-based visual servoing (Summer 2024). Revived legacy codebases for the lab’s RTVS and Imagine2Servo methods and built a benchmark evaluating them across varying viewpoint angles and query-to-goal distances.

B.Tech + MS by Research, Dual Degree (CSD)

International Institute of Information Technology, Hyderabad

Computer Science

• CGPA: 8.04 / 10
• Currently in the 4th year (semester 7). Expected graduation May 2028.

Plann3r: Predicting Planning Costs Grounded in 3D

Conference on Robot Learning (CoRL)

A. Vadali, K. Pandya, V. Garg, N. Chandna, D. Craggs, F. Dayoub, K. M. Krishna, S. Garg.
First author. Accepted; presenting in person at CoRL 2026, Austin, TX, USA (Nov 9–12, 2026).

PixelLoop: Shortcut Topological Navigation with Pixel-Level Loops

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

S. Chittawar, V. Garg, A. Vadali, K. Pandya, R. Jayanti, S. Garg, K. M. Krishna.
Accepted.

Most Impactful Project — HackIIIT 2026

IIIT Hyderabad

Mess crowd-prediction system using live QR-scan data and ARMA forecasting to recommend visit times; integrated into the official MyIIIT app and used daily by 1000+ students.

3rd Place — Lloyds Banking Group Hackathon

Lloyds Banking Group IP&I

Agentic AI system on Google Cloud that detects and corrects data inconsistencies with a human in the loop.

1st Place — Megathon 2025, IIIT Hyderabad

IIIT Hyderabad (Qualcomm-sponsored)

SnapGen: a prompt-driven Stable Diffusion style-transfer editor, deployed fully on-device on the Qualcomm QIDK NPU.

2nd Place — Megathon 2024, IIIT Hyderabad

IIIT Hyderabad (Qualcomm-sponsored)

Yoga instructor app: on-device human keypoint detection on the Qualcomm QIDK NPU, with real-time pose-correction feedback derived from joint angles.

Splatt3r-SfM — 3D Vision

Combined MASt3R-SfM N-view pose estimation with Splatt3r’s 2-view Gaussian head, fusing pairwise splats into a single aligned scene for feed-forward novel view synthesis from unposed images.

NoMaD Deployment on Pioneer P3DX — Robot Navigation

Deployed an image-goal navigation model on a Pioneer P3DX over ROS for closed-loop control, optimizing inference for real-time onboard operation.

Ray Tracer in C++ — Computer Graphics

Full ray-tracing pipeline with analytic intersection for five primitive types, plus lighting, shading and shadows, rendered via OpenGL.