Skip to content
CuriousDevs

Physical AI Lab

Intelligence that moves matter.

Client work keeps the studio running. This is our long bet: an intelligence layer that turns model output into controlled action on real machines, and the body it runs in.
Intelligence
OJAS
Embodiment
PARTH
Target hardware
Existing edge compute
Status
Early engineering

01 · Why this layer

A model can describe a scene. It cannot decide whether acting on it is permitted right now.

Model output carries no notion of what is legal, safe or currently possible, and no memory of what the machine was doing a second ago. Everything between the model and the motors is the engineering problem — and that is where we work.
Line illustration of the perceive, understand, decide and act loop around the OJAS core, with decide highlighted

Perceive

Sensor streams become structured observation: detections, poses and measurements the system can reason over.

Understand

Observations fuse into one live picture of the scene and the machine, held alongside memory of the task.

Decide

The model proposes. Explicit rules, permissions and constraints decide whether that proposal may execute.

Act

Authorised intent reaches real devices through stable interfaces, with independent safety hardware underneath.

02 · Brain and body

One decides. One moves.

The boundary between intelligence and embodiment is deliberate and typed. Intent flows down as structured requests; state flows up as structured feedback. Each side can evolve on its own clock.

Intelligence · The brain

OJAS

A general-purpose execution runtime for Physical AI, in early engineering. It is designed to connect models to cameras, sensors and machines through adapters, and to put a policy check with no machine learning in it between every proposal and every actuator.

  • Designed to run a CuriousDevs model, a third-party model or your own — behind one contract
  • Designed so that every action passes an explicit policy check
  • Planned adapters to ROS 2, simulation and other backends
  • Targeting Jetson-class, Raspberry Pi and x86 GPU hardware

Embodiment · The body · Paused

PARTH

A human-scale robot designed to bring OJAS into the physical world. PARTH is a customer of OJAS, not its owner — the same runtime is meant to run machines we did not build. Work on PARTH is paused until OJAS can drive a machine end to end.

  • Designed with independent safety hardware below every software layer
  • Stationary, wheeled and bipedal configurations planned on one stack
  • Commercial actuators, compute and sensors before anything custom

03 · The stack

Model-agnostic. Policy-gated. On hardware that already exists.

Open a layer to see what it does and who owns it.

Perception, world state, reasoning, planning, memory, rules and policies, tools and actions, telemetry and a device abstraction — the execution system between the model and the machine.

Ownership

CuriousDevs Intelligence

INTENT FLOWS DOWN · STATE FLOWS UP · OJAS RUNTIME OWNS THE MODEL-TO-ACTION PATH

04 · Where it stands

What exists, and what does not yet.

Built in order: runtime first, then the data and training platform, then our own model. Each row says what it is today, not what we hope it becomes.
ComponentStatusNext step
OJAS Runtime v0.1In engineering — architecture firstCanonical interfaces, runtime core and a first adapter
Backend adaptersDesignProve portability with a second backend
Data and training platformResearchBegins once the runtime is stable
OJAS ModelResearchStart from open VLA models, then train our own
PARTHPausedResumes when OJAS can drive a machine end to end

STATUS AS OF OCTOBER 2026

Founding team

Build OJAS with us.

We are forming a small founding team of two or three engineers for the runtime: robotics middleware, real-time systems and edge inference. Send the hardest thing you have shipped.

Write to us

Partners and early users

Have a machine, a task and a tolerance?

Describe the station, not the vision: the machine, the task, and what happens today when it fails. Research partners, labs and grant programmes are welcome too.

Talk to the lab