FIG. 01 / A HEADSET FOR PHYSICAL EXPERIENCE
We’re building a sensory intelligence layer that brings task-relevant signals, body state and prior experience onto a shared timeline, helping physical AI understand what is happening and anticipate what comes next.
CURRENT WORK / SIMULATION-BASED RESEARCH
Select an input or derived state.
FIG. 02 / HAND-WORN SENSING
The head observes the scene. The hand records contact, tactile patterns and force. Thermal sensing and chemical sampling are task-specific additions, used when the activity calls for them.
AS–002 / ASSEMBLED HAND / PROPOSED CAPTURE CONCEPT
HEAD + HAND + TASK CONTEXT → ONE SHARED EXPERIENCE TIMELINE
FIG. 03 / EVIDENCE → ACTION → OUTCOME
SCROLL TO FOLLOW · OR SELECT A MOMENT
01 / APPROACH
Scripted illustration. Synthetic signals in arbitrary units, illustrative timing; no experimental result or safety assessment.
FIG. 04 / A MOMENT, PRESERVED
A shared timeline connects task context, sensory evidence, body state, prior experience, actions and outcomes. Timing uncertainty and missing observations belong in the record too.
FIG. 05 / ONE CONNECTED SYSTEM
A head-worn unit captures scene, sound and inertial cues. Hand-worn sensors capture tactile contact, touch and force, with thermal or chemical modules for selected tasks. Their experience records feed the same training and evaluation loop.
Headset: RGB cameras, microphones and an IMU with gyroscope and accelerometer. Hand: tactile contact, touch, force and local motion, with thermal sensing or chemical sampling for selected tasks. Whole-body balance would also require body-state and support-contact evidence.
PROPOSED HARDWAREPreserve native timestamps, uncertainty and gaps. Link what was available to the action taken and the outcome that followed.
RECORD ARCHITECTURECompare matched representations and model capacity in simulation. Test held-out environments and timing disruption. Physical testing remains a future validation step.
CURRENT / SIMULATIONOUTCOMES → EXPERIENCE HISTORY → THE NEXT TRAINING AND EVALUATION CYCLE
DIRECTION / SIMULATION-BASED RESEARCH
We are studying sensory ordering and task-relevant evidence in simulation. No validated task-learning or physical-transfer gain is claimed. The scenes above explain the research direction and are separate from those experiments.
Does connecting senses, body state and prior experience help physical AI understand events and anticipate their consequences in unfamiliar real-world settings? That is the ambition to test, not a demonstrated real-world result.
Integrated capture hardware, measured synchronization in physical tasks and transfer to real-world settings remain future validation steps. Better prediction, safer handling and improved balance are research goals; this site claims no demonstrated improvements in those capabilities.
Evaluation should use matched model capacity, independent seeds, held-out environments and controls that disrupt timing while retaining the same channels. Sensor lag, uncertainty and missingness must remain explicit.
ABOUT US / AISTHE
Aisthe takes its name from koinē aísthēsis, the faculty that brings separate senses into a single perception.
We’re building toward a sensory intelligence layer for physical AI: task-relevant evidence, body state and prior experience connected to actions and outcomes. Our current research is simulation-based. Integrated capture hardware and real-world validation remain goals.
THE TEAM
Chief Executive Officer
Develops the sensory-priors thesis and the falsification discipline behind the research. Previously built FERAL, a local-first runtime with long-lived memory for physical AI agents. Writes the code.
GET IN TOUCH
Research groups, potential capture partners and robotics teams exploring task-relevant sensory experience.