Salticid Motion Lab

Flat-floor laboratory

Physics + controller

Actuation source

Canonical testing mode100% source targets · network, hybrid actuator transform, reflex, and learning bypassed

Canonical examplePlayback speed

speed 0.00 m/sheight 0.00 mupright 1.00feet 0/8forward/lateral -0.02 / 0.00 mheading Δ 0.0°elapsed 0.0 s

Use edited canonical clipBuilt-in analytic source is active

Training trials, rewards, optimizer updates, and model weights do not advance in this mode.

Control architectureReward architectureTraining populationElite population sizeNovelty weightRandom elite restart rateCoupled oscillator CPGProduces coordinated phase and stride modulation.Fast reflex layerLevels the body and responds to slip or instability.Canonical motion guidanceOptional teacher; validate cycles in source-test mode before enabling.Canonical guide strengthGuide fade · network updatesEpisode domain randomizationUses the versioned physical ranges stored in the checkpoint.

Version-4 checkpoints load in compatibility mode. New models use the modular research stack and persist every setting in version 5.

Enable motion clockProvides periodic features; no leg sequence is prescribed.Physical auto-tuneRaises motor force from measured tracking error.Render shadowsOff by default for training; physics and learning are unchanged.Dashboard refreshMotor strength90%Grip100%Latent clock frequency0.72 HzContact feedback0.85Gravity9.81 m/s²Simulation rate1.00×Abdomen bias0°

Static neutral-angle trim added to the network's learned abdomen target. It changes posture, not the learning signal.

Human reinforcementTRIAL TOAL +0

Each click changes this trial's optimizer score. Next trial applies the accrued score immediately; Reset discards it.

Position + heading previewpreview

Average velocity toward goal0.000 m/sHeading alignment0.0%

Emergent coordination

Live leg activation timeline

Rolling history updates continuously. Warm material = positive activation; cool material = negative activation.

Leg

-15

-14

-13

-12

-11

-10

-9

-8

-7

-6

-5

-4

-3

-2

-1

now

Higher-order body activations

Left

0.00

Right

0.00

Front

0.00

Rear

0.00

Pair 1

0.00

Pair 2

0.00

Pair 3

0.00

Pair 4

0.00

Segments, thorax, abdomen, pedipalps / papillae

Segment 1

0.00

Segment 2

0.00

Segment 3

0.00

Segment 4

0.00

Segment 5

0.00

Thorax pitch

0.00

Thorax yaw

0.00

Thorax roll

0.00

Abdomen

0.00

Palp 1

0.00

Palp 2

0.00

Palp 3

0.00

Palp 4

0.00

Palp 5

0.00

Palp 6

0.00

Structured training program

Turn toward destination

stage 1/4

Each learning update samples a new destination. Its heading is derived from the ray from the robot's initial pose to that destination, so matched +/− probes learn to orient first and then move forward. Timeout, success, or a fall completes the current probe.

Objective source

Position goal X0.0 mPosition goal Z-2.2 mHeading goal0°

Average velocity toward goal0.000 m/s

Heading alignment0.0%

Trial0

Best forward 20 m—

Network updates0

Reward0.000

Loss0.000

Goal X / Z0.00 / -2.20 m

Goal heading0.0°

Objective #0

Goal distance0.00 m

Heading error0.0°

Feet down0/8

Render FPS0.0

Physics Hz0.0

Worker snapshots0.0 Hz

Dashboard updates0.0 Hz

Draw calls0

Triangles0

Training heat1.00×

Upright1.00

Stage trial progress0%

Success rate 0% · trial 0.0/9 s · last reset

Training trials will draw reward and loss here.

RewardLossPosition velocity, heading alignment, stability, and motor tracking.

Research diagnostics

Live locomotion observables

Reward0.00

Velocity0.00 m/s

Heading alignment0.0%

Center of mass height0.00 m

Slip distance0.00 m

Energy consumption0.00

Stride frequency0.00 Hz

Stride length0.00 m

Duty factor0.0%

Novelty score0.00

Episode duration0.0 s

Recovery events0

Actuator utilization0.0%

Human feedback0

Footfall sequence—

Behavior stateidle

Elite lineagecompatibility baseline · 0 elites

Randomized domainμ 1.15 · g×1.00 · m×1.00

Human feedback this trial+0 · 0 reward · 0 loss

Canonical source testoff · 0% authority · learning off

idle

walk

run

turn

rotate

climb

recover

jump

forwardProgress 0.000headingAlignment 0.000stability 0.000posture 0.000energyEfficiency 0.000recovery 0.000obstacleSuccess 0.000livingCost 0.000legacy 0.000potential 0.000guidance 0.000humanFeedback 0.000

salticid-hierarchical-equivariant-v2checkpoint v51137 parametersmodular-cpgpotentialelite-spsa

No training session active