Motor Intelligence Infrastructure for Physical AI
We build the execution layer that makes physical AI systems actually work, across changing environments, with safety guarantees, personalized to each user.
Under the hood: state-aware control, objective inference from expert behavior, and hard safety limits, so systems keep working when the body, environment, or dynamics change.
Physical AI has a transfer problem
Physical AI can look impressive in a controlled setting, until the dynamics change.
Different bodies. Different environments. Different constraints. Small shifts in dynamics create big failures.
The gap is not more AI. It is the missing execution layer: a system that adapts actions in real time to the dynamics that actually exist, while staying safe and auditable.
RAST, the execution layer
RAST is a software stack that runs alongside an existing control system. It converts expert demonstrations into a compact intent package, then computes actions continuously from live state while enforcing hard safety limits.
Offline tooling that infers the motor objective behind expert behavior from demonstrations, and exports a deployable intent package.
On-device execution that recomputes actions from live state and intent, staying aligned as operating conditions shift instead of replaying a brittle policy.
Deterministic constraint enforcement. Clamps unsafe actions, monitors execution, and produces audit logs for safety-critical use.
Not a policy library. A state aware execution loop with hard constraint projection.
Cardiac rehab first. Built to expand.
We start with cardiac rehabilitation to validate outcomes, workflow fit, and safety constraints. The same execution layer extends to other domains where conditions shift and correctness matters.
Objective baseline, dose aware progression, and payer ready reporting. State aware guidance with hard safety limits, designed for clinic workflows.
Personalized movement change is not just harder exercises. RAST models each person’s movement strategy and guides minimal, safe adjustments that can be measured session by session.
When latency or environments break direct control, send intent and let the system execute locally under current conditions, inside explicit constraints.
State-aware control, in the loop
- State-aware action computation: actions are computed continuously from live state, not selected from a static script.
- Objective inference from experts: infer what experts optimize for, not only what they did.
- Minimal-change corrections: adjust only what matters to the task, preserve the rest of the natural strategy.
- Hard safety limits: bounds enforced in the loop with auditability and predictable behavior.
We do not claim magic. We engineer the bridge from mathematical movement models to deployable systems.
Team
I have spent decades building high-reliability systems where correctness and operational reality matter. Today I apply that mindset to Physical AI, turning rigorous movement science into software that can run in the loop, safely and measurably, in the real world.
I am an MSc in Autonomous System and Robotics from the Technion, focused on Deep RL and Control Theory in Complex environments. I bridge scientific rigor with practical implementation, drawing on R&D experience at Rafael and The Technion advanced robotic lab.
- Technion BCI Lab - validation of motor intelligence inference
- Clinical partner in Israel - first deployment focus in a Cardiac Rehabilitation Center, with outcomes measurement
Contact
If you are building a system where intent meets dynamics - robotics, rehab, or safety-critical automation - let’s talk about deployment paths and pilot design.