Hi, I'm Saketh —
I build robots that think.
By day I build physical agents — teaching machines to see, move and make decisions, from factory floors to operating rooms. By night I'm sketching gadgets, strumming guitar, and re-watching the cartoons that got me into this in the first place.
I design and ship autonomous systems end-to-end — perception, planning and control. Lately that means teleoperation systems, manipulation policies trained in Isaac Lab, and digital twins that close the sim-to-real gap. I'm happiest where messy real-world sensor data meets clean, reliable code, from soldering a board to deploying a full ROS 2 stack.
A robot agent that chats with you through OpenAI LLMs wired into ROS, then mixes your drink. Runs on GCP with Pub/Sub for robot-state monitoring, MoveIt 2 + OctoMap for collision-aware pouring, and a FastAPI bridge to the outside world.
A vision-assisted robot that inspects screws to 0.01 mm precision. A Doosan A0509 over MODBUS, an OpenCV eigenvector trick to localise each screw, YOLOv8 for detection, and Gazebo + MoveIt for motion planning.
Manipulation policies (ACT, Diffusion Policy) trained entirely inside NVIDIA Isaac Lab with parallel, manager-based environments — then run for real. Part of the HIR teleoperation system.
A digital twin of our robot cloned into Isaac Sim, so we can generate synthetic training data instead of paying for expensive real-world collection.
Teaching a custom-designed dexterous robotic hand in-hand manipulation with PPO in Isaac Lab. Currently at a 64% success rate and climbing.
A quadruped robot (canis = Latin for dog) with its own legged-locomotion stack. Equal parts engineering and weekend obsession.
A robot arm that turns images into stroke plans and draws them — where motion planning meets a bit of creative expression.
A motion-capture algorithm that lets plain hand motion drive a robot — no gloves, no extra hardware, just a camera. Core project of the HIR team.
A sensor-packed glove that captures the operator's hand pose and streams it to the robot, making teleoperation as natural as putting on a glove.
A custom gripper with its own ROS packages for simulation in Gazebo Classic and a MoveIt Setup Assistant config for planning and control.
A safety layer for shared workspaces: when a person steps into the robot's space, it smoothly slows down or stops.
RMPFlow integrated with Isaac Sim and MoveIt 2 for smooth, reactive, collision-aware arm motion in cluttered scenes.
- Building physical agents — robots that learn in simulation and work in the real world
- Led the HIR (Human Intelligence Robotics) team: a motion-capture teleoperation algorithm where hand motion drives robots with no extra hardware
- Trained ACT and Diffusion Policy in Isaac Lab; integrated RMPFlow with Isaac Sim + MoveIt 2; built a digital twin for synthetic data
- Shipped Ask Alice, the LLM-powered robotic bartender, and ScrewQC, vision-guided screw inspection at 0.01 mm
- EtherCAT communication for motors and sensors; Beckhoff controllers and Harmonic drives
- Machine vision with TensorFlow + OpenCV, built to ISO 13485 / FDA requirements
- Designed an automatic fruit & vegetable sorter to cut 25–30% post-harvest loss
- 24 fps vision model grading size, shape, colour and ripeness; SolidWorks + Raspberry Pi + Arduino
- AR-based visual haptic modelling for thoracoscopic surgery training — surface reconstruction from CT/MRI scans with VTK, ITK and ARToolKit
My life is more like a Telugu film — full drama, half emotions, konchem comedy, inkoncham confusion. But somehow the hero still survives. I fell in love with robots because of cartoon gadgets, still doodle weird machines in notebooks, noodle on guitar when the solder fumes get thick, and rewatch Doraemon for the gadget-of-the-week inspiration.
Doraemon's gadget-of-the-week episodes are basically my product roadmap. Each one is a tiny speculative design brief.
Notebooks full of machines that will never (and should never) exist. The sketching is the point, not the building.
Strumming chords when the solder fumes get thick. Debugging a tricky riff hits different than debugging code.
Soldering small delightful robots that do one silly thing well. The sillier the better — utility is overrated.
- Re-watching Doraemon — always good for gadget inspiration
- Training a dexterous robot hand in Isaac Lab (64% success and climbing 🔥)
- Learning to draw expressive robot faces
- Getting better at guitar one chord at a time 🎸
Because the best engineers never stop being kids.
Every Doraemon episode runs the same loop: someone has a problem, a gadget comes out of the pocket, things go hilariously wrong, and everyone learns something. That's honestly my job description.
Kids ask "what if?" without worrying if it's a silly question. They take things apart just to see what's inside. They believe a robot can be your best friend. I try hard to never grow out of that — it's the part of me that actually builds things.
Robotics gig, a collab, or just to argue about Doraemon gadgets — I'm in.