Why Heylper

Why Heylper?

A robot has motor skills, not a procedure.

So it needs a planner.

The planners that exist reason, then plan — a different plan next run.

Deployed work needs the expert's plan, every time.

So the planner must remember, then plan — and hold the robot to it.

That is Heylper — the agentic planner.

1.A robot has motor skills, not a procedure — so it needs a planner

The job"Service the control cabinet."

↓

The planner — decides the next step, checks the last

Gemini Robotics-ERGoogle

Cosmos ReasonNVIDIA

Hi RobotPhysical Intelligence

Heylperthe Expert Hat

↓ one instruction at a time — by voice to a humanoid, by tool call on ROS

The robot's motor skillsA VLA (Helix, GR00T) or ROS tools. One instruction in, motion out.

A VLA does one instruction at a time; it does not know the job. Something above it has to supply the next instruction and check the last. Google, NVIDIA and Physical Intelligence each built that layer. Heylper sits in the same slot, beside them, speaking to the same robots. The question is only what the planner plans from.

2.Those planners reason their plans. Deployed work needs the expert's.

Other planners: reason, then planGiven a mission, the model works out a route. A reasonable one — and it may work out a different one next run.

Heylper: remember, then planThe route is the expert's — learned from their demonstration, refined with people, recalled at run time. The same every run.

↓

…and holds the robot to itEach step is checked against the expert's own definition of done — through the Hat's camera — before the next one is given. Off the procedure, it stops and asks.

Heylper still reasons — about whether the step is done, when to speak, how to recover. It reasons within the procedure, never about it. That is what makes it agentic and conformant at once.

3.So the planner must remember — and this is where the memory comes from

LearnThe expert demonstrates; the apprentice watches and captures the procedural skill.

→

RefineIt works alongside people, session after session; corrections sharpen it.

→

GuideIt does the job alone, and Heylper holds it to the procedure.

Motor skills are learned from teleoperation and refined in simulation. Procedural skills are learned from experts and refined with people. Heylper is the second pipeline.

That is Heylper: the agentic planner that lets robots repeat procedural skills like experts.

Explore Heylper

The reasoning behind putting determinism in the decision rather than the motion is in our research.

Back to home