Skip to content
CuriousDevs

Intelligence that
moves matter

Intelligence built to perceive, reason, and act in the physical world.

Closing the loop between sensing, decision, and motion.

Explore OJASMeet PARTH
Intelligence
OJAS — perception, world state, reasoning, planning, memory, policy and action.
Embodiment
PARTH — a human-scale, electrically actuated robot for indoor handling work.
Where it runs
Existing edge compute: Jetson class, Raspberry Pi, x86 with a GPU.
How we talk about it
Nothing is described as available before it has recorded, repeatable evidence.

Physical AI

Intelligence is only half of it. The other half is action.

A model can describe a scene. It cannot decide whether picking that object is permitted right now, hold a coherent picture of a workcell across minutes, or keep a machine safe when its own output is wrong. Everything between the model and the motors is the engineering problem, and that is where CuriousDevs works.

Perceive

Sensor streams become structured observation: detections, poses, 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.

How they fit

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. That is what lets the intelligence and the body evolve on their own clocks.

OJAS

INTELLIGENCE
  • Perception and world state
  • Reasoning, planning and memory
  • Rules, permissions, forbidden actions
  • Tools, actions and telemetry

PARTH

EMBODIMENT
  • Skills and motion planning
  • Whole-body real-time control
  • Actuators, sensors and power
  • Independent safety hardware

OJAS decides what to do. PARTH decides how the body moves. Models never drive motors directly, and the interface between the two is typed, versioned and auditable — which is why the intelligence and the body can evolve independently.

Why this layer exists

A trained model is not a system that can be trusted with a machine.

Model output is numbers. It carries no notion of what is legal, safe or currently possible, and it has no memory of what the machine was doing a second ago. In the physical world, a mistake is not a retry — it is a stopped machine, a scrapped part or a damaged tool.
Nothing authorises the output
Inference alone puts nothing between a proposal and an actuator. Authority has to live outside the model.
One-shot inference is not a loop
Acting in the world means perceiving, deciding, acting and observing continuously — not answering once.
Behaviour slips quietly
Light changes, a fixture moves, a finish changes. Without recorded outcomes, degradation has no alarm.
Every machine speaks differently
Without a device abstraction, no engineering carries from one robot, gripper or controller to the next.

So we build the execution system first: model-agnostic, policy-gated, observable, and running on hardware that already exists.

OJAS · intelligence

The intelligence system behind physical machines.

It perceives, plans, decides, and acts — at the edge.
  • A CuriousDevs model, a third-party model or your own — all behind one contract.
  • Every action passes a policy check that contains no machine learning.
  • Runs on existing edge compute, not on hardware we ask you to buy.
  • Ships as one package: model, rules, tools, actions, configuration, metadata.
Explore OJASInside the architecture

PARTH · embodiment

Intelligence, embodied.

A humanoid robot built to bring OJAS into the physical world.
  • Independent safety hardware sits below every software layer.
  • Stationary, wheeled and bipedal configurations share one stack.
  • Commercial actuators, compute and sensors before anything custom.
Meet PARTHRobotics engineering

Working with us

What we take on.

A small set of problems we can be held to. If a conventional integrator is the better answer for your task, we will say so in the first conversation.

OJAS on your machine

Bringing a task to a running loop: perception, policy, actions, commissioning and telemetry on hardware you already own.

Bring your own robot

You own the machine and need the intelligence layer. One device adapter, then the full execution system above it.

Safety and policy architecture

Explicit rules and an authorisation layer above independent emergency-stop and Safe Torque Off hardware.

Evaluation and monitoring

Measuring how a model behaves on your task and your floor, then watching for drift once it is running.

Research

We publish questions before answers.

We propose no new model architectures. Our open questions are about execution: how published models behave under real conditions, whether policy checks hold at loop rate, whether degradation is detectable early, and whether a device abstraction really makes the second machine cheaper.

Company

Two divisions, one stack, built in Noida.

CuriousDevs Intelligence builds OJAS. CuriousDevs Robotics builds PARTH. The company is deliberately small, the boundary between the two is a contract rather than an org chart, and we are explicit about what exists and what does not.
Read the open questionsAbout CuriousDevs

Contact

Have a machine, a task and a tolerance?

Tell us about the station — the machine, the task and what happens today when it fails. A founder reads every message.

Talk to CuriousDevs