Robotics & Computer Vision  ·  4+ Years Industry

Karteek Gandiboyina

MS Autonomy & Robotics, UIUC (GPA 3.89)  ·  gkarteek99@gmail.com

Currently building autonomy stacks for wildfire-suppression UAVs at Rational CyPhy Inc. Previously at Konica Minolta R&D, Tokyo. B.Tech. Electrical Engineering from IIT Kharagpur. Deep expertise in 3DGS, 6-DoF pose estimation, SLAM, and ROS-based autonomy — backed by an IROS 2026 publication and 2 filed patents.

IROS 2026 2 Patents Filed GPA 3.89 / 4.0 JAXA Collaboration
Karteek Gandiboyina

Work Experience

Rational CyPhy Inc. Feb 2026 – Present
Robotics Engineer
  • Architected a resilient autonomy stack for NASA-aligned wildfire-suppression UAVs on VOXL 2, targeting BVLOS operation in GPS-denied environments.
  • Fused Basalt VIO with GPS; designed seamless sensor failover for stable flight under degraded conditions.
  • Deployed real-time 3DGS for aerial environment reconstruction, delivering high-fidelity situational awareness to emergency responders.
  • Prototyped self-organizing multi-drone architectures for mission continuity during comms/GPS outages.
Coordinated Science Lab, UIUC — Prof. Sayan Mitra Sept 2025 – Jan 2026
Graduate Research Assistant
  • Trained a DreamerV3 World Model on Crazyflie FPV data for imagination-based policy training without direct environment interaction.
  • Built a 6-DoF pose estimation pipeline using 3DGS-generated synthetic datasets (10k frames) + fine-tuned U-Net; achieved sub-centimeter accuracy in sim-to-real transfer.
  • Developed actor-critic (PPO) policy within latent world model; increased waypoint completion 70% → 95%.
  • Designed min-snap trajectory planners with differential flatness, reducing planning time by 30%.
Konica Minolta — R&D Headquarters, Tokyo, Japan Jul 2021 – Aug 2024
R&D Robotics Engineer
  • Structured-light point cloud sensor achieving 0.99 IoU on 1 cm² items at 0.05 m under <10 Lux for bin-picking of reflective industrial components.
  • Auto-annotation pipeline (SAM, Detectron2, YOLOv7) reducing manual labeling time by >80%.
  • Language-conditioned multi-task RL (SAC + DAgger) with Georgia Tech; improved zero-shot task success by 200%.
  • Collaborated with JAXA on vision-based 6-DoF grasping for multi-limbed robots aboard the ISS.
  • Filed 2 patents in computer vision and robotic grasping (USPTO & JPO).
Philips Innovation Campus — Bangalore, India Apr 2020 – Jul 2020
Machine Learning Intern
  • 3D residual U-Net for pulmonary nodule detection on 888 CT scans (LUNA16); FROC score 0.914, surpassing baseline by 8%.

Education

University of Illinois Urbana-Champaign, USA Aug 2024 – Dec 2025
MS in Autonomy & Robotics — GPA: 3.89 / 4.0
Deep Learning for Graphs · Computer Vision · Safe Autonomy · Deep Generative Models
Indian Institute of Technology Kharagpur, India Aug 2017 – May 2021
B.Tech. in Electrical Engineering — GPA: 3.48 / 4.0
Deep Learning · Machine Learning · Embedded Systems · Control Systems

Projects

Drone Racing
Autonomous Drone Racing

MPC & PID spline tracker for autonomous gate navigation. NanoSAM keypoint detector for misaligned gate correction.

Race record: 50.13 s (Tier-1) · Gate error: 0.05 m
MPCPIDNeRF AirSimNanoSAM
Code
VLM4Autonomy
VLM4Autonomy

SAM2 + optical flow + VLMs for real-time object tracking and ego-vehicle motion estimation. YOLOv8 & SfM for spatial grounding.

SAM2YOLOv8Optical Flow SfMVLMs
Code
Stock Predictor
Volatility-Aware Stock Prediction

Transformer-VAE with RealNVP normalizing flows conditioned on VIX. Outperforms TimeGrad & GARCH in distributional similarity.

Wasserstein: 0.0629 · Cosine Sim: 0.9072
PyTorchVAE RealNVPTransformers
Code

Patents & Publications

Conference Paper
FalconTrack: Photorealistic Auto-Labeled Perception and Physics-Aware Vision-Based Aerial Tracking
IROS 2026

Unified 3DGS-based simulator automating 6-DoF pose and mask labeling (10k images in <20 min), enabling zero-shot sim-to-real aerial tracking with 100% success on high-speed trajectories.

Patent — USPTO
Learning Data Generation Support Device, Movement Controller, and Article Acquisition System
US: 20250336175

Automated data generation framework eliminating manual labeling bottlenecks for complex 3D pick-and-place robotics in bulk manufacturing.

Patent — JPO
Component Posture Information Acquiring Device and Posture Determination Method
JPO: 2023-060687

Vision-based sensing system using ROI determination and feature extraction for accurate component posture estimation during robotic pick-and-place.

Skills

Computer Vision
3D Gaussian Splatting NeRF SLAM 6-DoF Pose Estimation YOLOv7/v8 Detectron2 SAM / SAM2 Point Clouds Optical Flow SfM Visual Odometry
Robotics & Autonomy
ROS / ROS2 MPC PID Min-Snap Trajectory VIO (Basalt) Quadrotor Control Robotic Manipulation Bin Picking AirSim MuJoCo MetaWorld
Machine Learning
PyTorch JAX CNNs Transformers VAEs Diffusion Models GNNs DreamerV3 Actor-Critic RL SAC DAgger LLMs / VLMs / VLAs
Languages & Tools
Python C++ Docker Linux Git VOXL 2