About
Hi, I’m Varun Vaidhiya — a Full-Stack Robotics & AI Engineer building intelligent systems. My work focuses on Vision-Language-Action models, ROS 2, and high-performance edge GPU inference for autonomous vehicles and physical AI.
I’ve built end-to-end robotic software — from 3D perception and SLAM to Vision-Language-Action model training and edge GPU inference optimisation. I specialize in C++ for embedded Linux and IoT device management, as well as building deep learning pipelines on Jetson and RTX edge GPUs.
I’m currently building OhhO Robotics — one robot, one open engine (the rest is roadmap). Working today: the engine can talk to OmniBot’s mecanum base, run a simulator with no hardware attached, and record arm and base motion for training. I’m interested in working with companies building the intelligence layer for robotics, autonomous vehicles, and UAVs.
Experience
OhhO Robotics — Founder & Robotics Software Engineer
Oct 2024 – Present
Five AI — MSc Dissertation Researcher, Autonomous Vehicle Perception
Dec 2023 – Oct 2024
42Gears Mobility Systems — Senior Software Engineer, IoT & Embedded Linux
Mar 2023 – Jul 2023
Tata Consultancy Services — Systems Engineer
Feb 2021 – Sep 2022
Education
- MSc Smart, Connected and Autonomous Vehicles — University of Warwick, WMG (2024) Autonomous Vehicle Perception, Deep Learning for Robotics, Sensor Fusion & Calibration, Real-time Embedded Systems, Functional Safety (ISO 26262), GPU Computing & Optimisation
- BEng Electrical and Electronics Engineering — RMD Engineering College (2020)
What I Work On
- 3D Perception & Spatial Intelligence — Multi-camera Bird’s-Eye-View (BEV), RGB-D 3D SLAM (RTAB-Map), 3D occupancy / voxel mapping (OctoMap), 2D SLAM, 6-DoF object pose estimation, point-cloud processing, multi-sensor fusion (EKF), camera calibration (homography, RANSAC, pinhole intrinsics)
- Machine Learning & Embodied AI — PyTorch, Vision-Language Models (CLIP, SAM 2, DINO), Vision-Language-Action Models (OpenVLA, SmolVLA, piZero, Octo), imitation learning (behaviour cloning), reinforcement learning (PPO, Isaac Lab), distributed training (DDP/FSDP), Weights & Biases
- Edge Deployment & Inference Optimisation — TensorRT (FP16/INT8), ONNX Runtime, CUDA, INT8 calibration & quantisation, model compression/pruning, Nsight profiling, NVIDIA Jetson / RTX edge GPUs, ultra-low-latency inference
- Robotics & Simulation — ROS 2 (Jazzy), Nav2, TF2, real-time control, Gazebo, NVIDIA Isaac Sim / Isaac Lab, sim-to-real transfer, teleoperation
- Systems & Infrastructure — Embedded Linux, ARM Cortex-A, System design, IoT protocol integration (DDS, MAVLink, CAN, OPC UA), C++ (C++17/20), Python, CMake. (Supporting Stack: Kubernetes, Docker, TypeScript, Next.js)
GitHub Activity
Open Source Projects
| Project | Description | Stack |
|---|---|---|
| OmniBot | Open-source mecanum mobile manipulator (Yahboom base, SO-101 arm, Raspberry Pi 5, Meta Quest 3 teleop). Working today: the Apache-2.0 ohho-os engine (PyPI 1.1.2) talks to the base, runs a simulator with no hardware attached, and records arm and base motion for training. Roadmap: the wider OhhO console platform — the website consoles are prototypes with simulated data, and fleet, twin, compliance and security consoles are roadmap. | Python / ROS 2 / TypeScript / Next.js |
| ohho-sdk | The ohho-os Python package on PyPI; CI on Linux/macOS/Windows | Python |
| OhhO-VR | Meta Quest 3 teleop client over ROSBridge | Unity / C# |
| SAM2forAV | SAM2 model applied to autonomous vehicles | Jupyter / Python |
| AEB-Model-Based-Design | Automatic Emergency Braking model-based design | Simulink |
| perfetto_analysis_repo | Performance analysis scripts using Perfetto | Python |
| libfreenect2 | Fork with Raspberry Pi / ARM NEON and USB transfer fixes | C++ |
Stay Connected
Follow me on GitHub · Find me on PyPI · Connect on LinkedIn · Follow on X/Twitter · Subscribe on Substack · Watch on YouTube
© Varun Vaidhiya