Every task, every embodiment, all at once.

In our previous blogs, we announced our mission of bringing physical AI to existing assets in production built with copper-rs, an open-source Physical AI OS. In this blog, we share our embodiment-agnostic hardware design philosophy central to Spatial Intelligence Kit (SIK), that made this mission possible.

But first: Why does SIK exist at all ?

Tens of millions of machines are already in production and deployed across a vast swath of industry verticals, but there is one problem. They lack intelligence. They can’t sense, think, and act autonomously. These assets need a human operator, f.ex pallet jacks, forklifts, wheel loaders, street cleaning vehicles, class B vehicles, and port terminal trucks.

Our first thought: Let’s just add intelligence.

On a closer look, a hard requirement surfaced: We need embodiment-agnostic hardware that serves as both as a data-collection and inference silicon, operating in the field, in diverse conditions, in power-constrained environments, and with limited connectivity.

This led us down the path of crafting a field-tested Spatial Intelligence Kit (SIK). SIK is already deployed on 100 assets in 30 locations, streaming multimodal data from the target environment continuously, with over 810,000 hours of operational uptime.

The Spatial Intelligence Kit can be paired with any physical asset in a couple of hours, after which the asset can be operated normally in the field.

When the asset is operated normally by operators or drivers, SIK captures rich multimodal data in the field, and especially in the target environment where it will be deployed as an autonomous worker. The multi-cam setup captures rich visual data, high-precision GNSS captures positioning data, and it reads actuation data through our custom wire harness connected to all of the CAN interface(s) and control interfaces.

SIK logs the actuation data from the asset during the operations. These are the operator’s inputs on the physical asset, f.ex throttle, steering, vertical fork position, excavator arm movement. In the background, our spatial intelligence learns the task from the multimodal data, f.ex moving pallets, digging in quarries, cleaning streets, and driving passengers. It pings the operator when it's ready to join the fleet as an autonomous worker.

Every task, every embodiment, all at once.