01
Perception
Turning cameras, sensors and physical signals into structured observations.
CuriousDevs Intelligence
Building the intelligence layer for Physical AI.
We build the models, systems and runtimes that enable machines to perceive, understand, reason, plan and act in the physical world.
From learned models to the systems that make intelligence work in the physical world.
What intelligence means
From seeing the world to acting in it.
Physical AI needs more than a trained model.
It needs systems that can interpret the physical world, maintain context, reason over changing conditions, make decisions, and turn those decisions into reliable action.
CuriousDevs Intelligence exists to build that layer.
The intelligence stack
Intelligence is a system, not a single model.
We work across the technologies that carry a machine from raw physical signals to intelligent behaviour. These are areas of work, not a list of released products.
02
World understanding
Maintaining representations of the environment, objects, state and change.
03
Reasoning
Understanding situations, goals, relationships and constraints.
04
Planning
Turning reasoning into structured sequences of actions.
05
Memory
Retaining context, experience and relevant state across time.
06
Models
Developing and adapting models for perception, reasoning, multimodal understanding, prediction and action.
07
Runtime systems
Allowing intelligence to operate continuously, in real time, close to the machine.
08
Evaluation and safety
Testing behaviour, measuring failure, constraining decisions and understanding system performance.
Models and systems
A model is a component. The system is the intelligence.
Models provide capabilities. The intelligence layer gives those capabilities context, memory, reasoning, planning, constraints, execution and feedback. We research and fine-tune models, work with external ones, and are building our own — and we build the systems that make any of them useful on a machine.
System intelligence
THE LAYER AROUND THE MODELModel intelligence
ONE COMPONENT
- Training
- Fine-tuning
- Multimodal models
- Perception
- Reasoning
- Action models
- Evaluation
- World state
- Memory
- Planning
- Runtime execution
- Tools
- Policies
- Feedback
- Real-time operation
Together, they turn model capability into physical intelligence.
The physical intelligence loop
Perceive. Understand. Reason. Plan. Act.
Physical intelligence is not a one-shot answer. The system observes the world, reasons over what it finds, acts, and then uses the result to inform what it does next.
01
Perceive
Signals → observation
02
Understand
Observation → state
03
Reason
Goals and constraints
04
Plan
Reasoning → intent
05
Act
Intent → machine
06
Observe
The result, measured
07
Update
State carries forward
RETURNS TO PERCEIVE
Design principles
Built for the physical world.
Intelligence that drives a machine has to hold up in conditions a digital-only model never meets.
- Continuous
- Intelligence operates as a loop rather than a one-shot response.
- Contextual
- Decisions are grounded in current world state and previous observations.
- Constrained
- Actions operate within explicit rules, permissions and physical boundaries.
- Feedback-driven
- Every action changes the environment and becomes information for what happens next.
- Edge-native
- Intelligence can operate close to the machine, where latency, availability and compute constraints matter.
OJAS is where this starts.
The models, world models, runtimes and evaluation systems that follow come from the same foundation. The division is built to be broader than any one of them.