Sai Navaneet
I graduated from Kyungpook National University with a Master’s in Electronics and Electrical Engineering, focused on automation and robotics. My background is rooted in electronic systems, control, and embedded technologies, developed through hands-on research and real-world robotic deployments.
My work centers on robotic manipulation and autonomous systems, where I build robotics software and learning-based models that bridge perception, decision-making, and control. I enjoy turning complex ideas into reliable, deployable robots that advance intelligent automation.
Latest Updates
-
Submitted to IEEE Transactions on Cybernetics
RL-based prescribed-performance neuro-optimal control for robot manipulators under composite actuator faults.
-
Submitted SUREFlow to IROS 2026
State-space uncertainty-aware residual flow matching · 92.6% LIBERO success rate with 179M parameters.
-
Building Tensiq TT · sensor-agnostic tactile pipeline
Working on the Tensiq tokenizer + TENSIQ CLI — turning raw tactile-sensor frames into standardized Tensiq Tensors (TT) for multimodal foundation models.
-
Promoted at Airobotics · Research → Senior Robotic Engineer
Yaskawa-driven weld bead detection on the vehicle assembly line · inline QA & robotic guidance models for autonomous car manufacturing.
-
MambaVLA published at CCNC 2026
Scalable state-space VLA transformer · IEEE Consumer Communications & Networking.
-
Oral presentation · MambaVLA at CCNC 2026 · Las Vegas
Presented MambaVLA — a scalable, low-latency VLA framework on Mamba state-space models — at IEEE CCNC 2026 in Las Vegas. Linear-time, millisecond-level inference for real-time robot deployment.
-
MambaVLA accepted at CCNC 2026
A scalable, efficient vision-language-action model built on state-space architecture.
-
Submitted to Neural Networks journal
DiffDAIL — diffusion-enhanced vision-guided imitation learning with discrete latent representations.
-
Poster at IROS 2025
LegMamba — a scalable, efficient state-space model for quadrupedal locomotion.
-
Building VLA models with state-space backbones
Ongoing work on MAMBA-style sequence modeling for vision-language-action policies.
-
Paper submitted to CCNC / CES 2026
-
Won LeRobot (Hugging Face) Hackathon · Daegu
First place in the imitation-learning track.
-
Submitted to NeurIPS
QROOT — integrated diffusion transformer and reinforcement learning for quadrupedal locomotion.
-
Submitted to Engineering Applications of AI
Vision-guided predictive action imitation learning with discrete latent encoding for multitasking robots.
-
Paper published at NODYCON 2025
DLDMP — discrete latent diffusion motion planning for manipulators.
-
Automated tissue processing with transformers
Built at Dexweaver Company.
-
Autonomous harvesting via YOLOv11 object detection
Field-deployed pipeline at Dexweaver Company.
-
Paper published in KNU-EERC 2024
Leader-follower robot tracking using model predictive control.