phydata
Research

A common trace for tools, hands, and machines

phydata ResearchData infrastructure

Agriculture, surgery, logistics, and skilled trades use different tools, but their data shares a common problem. Observations, actions, instructions, and outcomes arrive through different systems and at different rates.

A common trace makes those systems comparable without flattening the details that make each field unique.

One timeline, many streams

The trace stores each signal in its native form while connecting it to a shared timeline. Video, pose, force, tool telemetry, spoken instruction, and annotations can stay separate and still be queried as one demonstration.

Figure 1. One shared clock keeps every sensor and decision usable as a single training trace.

Keep raw data and decisions together

Derived labels should point back to the source evidence and the review decision that produced them. This makes corrections traceable and allows teams to rebuild training views when a schema or model objective changes.

Figure 2. Every training view remains linked to the evidence and review decisions that created it.

Interfaces should survive new hardware

The physical world will keep producing new sensors and machines. A durable trace model defines stable concepts such as time, frame, actor, object, action, and outcome while allowing new modalities to attach without rewriting the whole dataset.

Figure 3. New sensors can be added without rebuilding the core trace model.

Figures use illustrative values.

Tell us the capability your model is missing.

Describe the task, the setting, and what success looks like. We design the collection around it and deliver the data in your schema.