About me.
I'm Mikhail. I work on robot learning — teaching machines to move, grasp, and eventually to feel what they're holding — and I ship software people can actually download and run.
The thread through all of it is that I like owning the whole thing: the robot and the training pipeline, the app and the protocol it speaks. Click anything below to open it up.
the engine — Caliper
An open robotics engine in Rust: kinematics, IK, dynamics, contact simulation, dataset tooling and training verdicts in one small deterministic binary, with a desktop app, a CLI and a Python face. Built on the bet that most arm work needs neither a datacenter nor the ROS junk drawer. The page · the repo ↗
the research — ScaffDiff, HapticVLA, Phantom
Dense 3D scenes completed from sparse LiDAR in a single denoising step; robots that handle eggs without tactile sensors by predicting the touch they no longer feel; and a world model that calls the slip before it happens. Demos of all three live on /research.
the apps — Wisp, ScanPatch
the rack — two Raspberry Pis
Almost everything I use daily runs on two Raspberry Pis — LLM-triaged notes, agent orchestration, nightly off-host backups. The tour.
How I work, in one rule: ship the running version, then make it honest — measure it, break it, write down why it broke. I trust numbers over demos and demos over slides.
contact
One paragraph is plenty. If you want to work together, describe the outcome you're after — ranges and uncertainty welcome. If you have a question, tell me what you tried, what you got, and what you expected; that format gets the fast, useful answer.