Physical AI Laboratory
WIRES | CABLES | PLUGS | SOCKETS
Laboratory provides infrastructure to obtain ground truth physical measurements, behavioral videos and demonstrations of cables, wires, plugs and sockets. These measurements, behavioral captures and demonstrations are registered and read by computer vision. Derived parameters, traces and 3D / 4D shape data create a foundation of physically grounded measurements and data for physical AI. Use cases of such data range from simulation and validation in robotics to physically enabled world models.
OpenUSD, Isaac Sim, Replicator, Cosmos, Isaac Lab, Newton, MuJoCo
Computer vision extraction & analysis
All measurement videos, behavioral videos and demonstration videos are recorded with April tags and QR codes. These are used for a) components recognition and management b) camera scale and registration c) clapperboard readout and scene management d) tracking of the motion of hands and components e) tracking of the instruments and their readings.
Such recordings processed by computer vision remove human error, automate process, provide traceable archivable data and can be used for AI training.
Ground truth measurement
Readings can be obtained by a) tracking a point position in space, for example cable end b) directly tracking and reading instrument display with LLM. Readings come in position-time or instrument reading-time curves and are processed for each station. The derived parameters are computed based on physics equations.
For complete set of equations visit Measurements →
Parameters table
Station A — Basic metrology
overall diameter
mass / m
min bend radius
rest shape
rest curvature
coil memory
free length
ovality
inner composition
bend plane
rest twist
bend set
spring-back ratio
bend memory
damage bend radius
Station B — Cantilever
bend stiffness 0°
bend stiffness 90°
bend damping ratio
bend anisotropy
nonlinear bend stiffness
load-bend stiffness
residual bend
gravity sag fit
Station C — TVL Axial
axial stiffness
normal stiffness
axial hysteresis
compression hysteresis
axial damping
flattening limit
Station C - TVL Friction
static friction
dynamic friction
stick-slip
rolling/sliding mode
wear-in drift
Station C - TVL Mating
insertion peak force
extraction peak force
force-displacement curve
force canopy
locking mechanism forces
retention force
max lateral offset
jam force
Behavioral videos & validation
Behavioral videos represent visual cues how an asset behaves under dynamic movements. Video is registered and scaled using April tags. End cable tags with gloves tags are tracked and movement traces are extracted. The cable and approximate environment is manually modeled in simulation engine, such as Nvidia Newton. Movements are applied to the modelled cable. Such simulation plausibility is then compared to the original video behavior.
See behavioral videos and validation in Measurements →
Behavioral videos description
S - Spring
Cable is in a loop on the table. One end of the cable is picked up and moved up-down in a vertical fashion. Movement is repeated three times with varying speeds from slow, then moderate to fast.
K - Kinematics
Cable is in a loop on the table. One end of the cable is picked up and moved left-right in a horizontal fashion. Movement is repeated three times with varying speeds from slow, then moderate to fast.
H - Hands 1
Cable is in a loop on the table. Both hands grip whole spool from left and right. Spool is counter rotated with hands to form ∞ (infinite loop) and back.
H - Hands 2
Cable is in a loop on the table. Both hands grip whole spool from left and right. Spool is pulled and compressed towards the center to form ovals.
H - Hands 3
Cable is in a loop on the table. Cable in chaotically untangled to form spaghetti. These are then squeezed and released by hands multiple times.
H - Hands 4
Cable is untangled to start to form linear object. Cable is rounded by hands pitching it. Various parts of the cable and radiuses of the cable are performed.
USD Asset creation
3D geometry of asset is obtained by a) 3D modeling from XYZ photography b) available 3D CAD c) CT scanning. Physical measurements obtained before are attached to the asset creating a high quality representation of cable, wire, plug and socket. Library containt photos of inner wires as well.
See XYZ photography and download USD assets in Measurements →
Demonstration with 3D metadata
Setup is specifically adjusted to enable 3D metadata extraction by a) colored tape bands along the cable length b) spray painted texture c) Lidar depth recording d) TBA UV spray paint with UV light e) TBA OSMO tactile glove . These multimodal demonstrations are used as inputs for AI model training for example for cable-state video prediction model.
Demonstrations are not yet available. TBA
