DOOR OPENING
MEASURED / 20 Hz
—°remaining
START5.55 s
THE DECISION LAB
/ ROBOTICS
Small decisions.
Physical outcomes.
Watch a robot move.
See what Jev chose.
Loading the recorded episode…
THE MOTIONREADY
PANDA / OSC000 / 111
EXECUTING—
FREE SPACE
Close the microwave
20 Hz / SIMOriginal LIBERO success criterion.
INSIDE THE DECISION
01 / 14
03 / MOTOR
Which input?
—confidence
CHOSEN INPUT / TASK PROGRESS
FORECAST · ACTION END—
OBSERVED · THIS FRAME—
CHOICE / Jev— ms
00.00SIMULATION TIME05.55 s
DECISION TRACKCLICK TO EXPLORE
THE ACTION SPACE
27 → —
ChosenOffered to JevOther feasible
THE TRAJECTORY
Boundary samples · dashed: preview—
A LOOK UNDER THE HOOD
Explore the architecture ↗
From state to movement.
Preview the possibilities.
MuJoCo simulates short branches. Geometry and task conditions turn physical effects into a set of candidate inputs.
PHYSICS → CANDIDATES
Let Jev choose.
Three connected choices narrow the next move: the objective, the kind of interaction, and one atomic input.
INTENT → STRATEGY → MOTOR
Move. Observe. Repeat.
The robot executes the selected input. A fresh observation starts the next cycle. Each choice and control is recorded.
ONE INPUT → NEW STATE
YOUR NEXT EXPERIMENT
Same engine.
Same engine.
Your task.
Start with a bundled task, then shape the goals and decision criteria in JSON.
GET STARTED
pip install -e '.[robot]'
jev-libero run --provider typesafe \
--task top_drawer --out runs/demo