Concept visualization: Luna and Sharpa robots beside an open dishwasher; Luna holds a plate above the rack.

InvestigativePhysicalIntelligence

Built on our own simulation technology,
from physics solvers to rendering.

Scroll to explore

The simulation foundation

A world built
from the physics up.

One system for physical interaction and visual experience.

01Physics solvers
Model the interaction.
Start with the physics. Our in-house solvers provide the foundation for how objects move, make contact, and respond.
02Rendering
See the same world.
Our own rendering brings that simulated world into view. Physical interaction and visual experience belong to the same system.
03One simulation foundation
From physics to perception.
Built in-house, from physics solvers to rendering. A foundation for reconstructing situations, exploring explanations, and making predictions to test.

Illustrative cut through an ivory material, showing blade contact and local deformation, without a particle mesh. — Physics solvers to rendering

Simulation makes predictions. Reality supplies the evidence.

Our approach

An unexpected failure
is evidence.

Illustrative investigation scene: a broad-faced ivory plate standing in a black dish rack. The same scene is used throughout the four reasoning steps.

What could explain the failure?

A plate will not fit.
The motion, the geometry, or the contact?

Motion path
Perceived geometry
Physical interaction

Inside the reasoning

Begin with the evidence. Form competing explanations rather than treating every failure as a reason to repeat the same action.

The inner loop · Investigate this situation

What would each explanation predict?

Reconstruct the plate, the rack, and their interaction. Examine what should change if an explanation is correct.

Reconstruct the scene
Compare predictions
Choose an intervention

Inside the reasoning

Use simulation to reason through consequences and compare possible interventions. A simulated outcome is a prediction to test, not proof about reality.

The inner loop · Investigate this situation

Which evidence separates the explanations?

Choose a real-world experiment that distinguishes a motion error from a mistaken model of the scene.

Design a real experiment
Gather evidence
Revise the explanation

Inside the reasoning

Bring predictions back to the physical world. Use the result to revise the explanation and the next action, working toward a successful outcome.

The inner loop · Investigate this situation

What carries forward to the next case?

Look across cases for knowledge and investigative methods that remain useful when the task or scene changes.

Compare across cases
Build reusable knowledge
Improve skills and tools

Inside the reasoning

The outer loop improves the investigator itself: its understanding, skills, and tools. Inductive learning can also refine a hypothesis within a single investigation.

The outer loop · Improve the investigator

The inner loop · Investigate this situationScroll to investigate · or select a stepSelect a step to explore

A closer look

Make the work
inspectable.

Explore task definitions, recorded model rollouts, and outcomes in our current benchmark.

Explore the benchmark

InvestigativePhysicalIntelligence

Model the interaction.

See the same world.

From physics to perception.

An unexpected failureis evidence.

What could explain the failure?

Make the workinspectable.

An unexpected failure
is evidence.

Make the work
inspectable.