ManiLoop Robot workspaceLANGUAGE → ACTION → OBSERVATION
← Project
Recorded interactive workspace

Look through the robot’s decisions.

Explore the real workspace with a complete, recorded GPT-6 episode.

Run your own task ↗
You are browsing a recording. Controls navigate saved observations; they do not call a model or run new physics. All 46 decisions come from one uninterrupted episode.
Simulation view
Saved camera frames
Recorded external camera before decision 1 EXTERNAL CAMERALIBERO / PANDA
Recorded wrist camera before decision 1WRIST CAMERA
XYZWORLD FRAME
0.00 s
TCP position · m
—
X / Y / Z
Gripper opening
—%
0 closed · 100 open
Decision
1/ 46
move
Simulation time
0.00s
Recorded sensors
Model’s action explanation
MOVE

The model’s stated interpretation, not an independent measurement.

01Episode playback
Paused

Pick up the black bowl between the plate and the ramekin and place it on the plate.

Single step alternates before/after views. Decision replay skips between recorded keyframes; the complete video below shows every control step.

02Recorded configuration
Read only

These are this episode’s settings. To change the task, model, or API connection, run ManiLoop locally.

Episode’s final result
SUCCESS

Official LIBERO success at control step 706. This is the episode’s final evaluation, not the score at the frame currently selected.

Success triggered during the final lowering action. The environment stopped with the gripper still closed; release and retreat were not tested.

Human review only · Evaluation was excluded from policy input.

Read the result and conditions ↗
Original field names
Execution feedback
All 46 decisions
Select any decision
The complete episode · every control step
707 FRAMES · 20 FPS

35.35 seconds of simulation playback, including all grasp retries in this single attempt. Model waiting time is omitted. No cuts, interpolation, or newly generated actions.

Use the live workspace on your computer

The local application lets you select tasks, configure model access, and execute new robot actions. Install the main and LIBERO environments, then run:

python -m maniloop demo --backend libero --codex-login --port 8767