Moravec's Paradox says that machines can do some amazing things that we humans find hard - and yet the things we learn as children, machines still struggle with:

How to avoid collisions.
How to read hesitation.
How to move with others.
How to sense what may happen next.

They're some of the fundamental human instincts that we use every time we go out into the world – they're the things that keep us safe.

Yet they remain some of the hardest things for Physical AI – robots and self-driving cars – to learn.

Perception tells a machine what is there.
Planning tells it what to do.

What's missing is understanding movement, intent and consequence – those instincts that keep humans safe, transferred into the realm of robotics and AI.

At SpatioTemporal, we're building a family of intelligences to address that gap:

SPATIAL – Motion, intent and human awareness.
TEMPORAL – Future-state predictions, causality and consequence.

We believe these are the missing middle stack Physical AI needs to interact safely around humans.

We're beginning with movement – where intent becomes visible.

We've built a foundation model that treats motion as a language, and intent as something that can be inferred. Not from pixels, but from how things move through space and time.

Our model can help infer signals like hesitation, distraction, instability, and changing intent – often before explicit actions occur.

Allowing robots to read the room, and cars to read the road.

In NVIDIA Cosmos simulations designed to place robots and humans in close proximity, we saw near-collisions fall by over 90% when we introduced our Motion Intelligence model. We saw:

Earlier yielding
Smoother shared-space negotiation
Less planner volatility.

This is early proof of the first layer: Motion Intelligence changing robot behaviour in human spaces. The broader opportunity is to build the full family of Spatial and Temporal Intelligences around it.

We also use these models inside simulation: varying not just weather, lighting and layouts, but human behaviour itself – so robots can train on hesitant, distracted and unpredictable people before meeting them in the real world.

The new era of Physical AI will need more than perception and planning. It will need machines that understand humans well enough to be trusted in human spaces.

That's the frontier we're exploring – from Motion Intelligence to a broader family of SpatioTemporal Intelligences.

If this is a world you need to understand, connect on LinkedIn.