Control real robot arms with AI. Watch, prompt, and run. API docs · Run locally · ASPIRE README
Paste this into your coding assistant
Clone https://github.com/andlyu/blupe-remote-yam.git and launch ./run-playground.sh. Help me connect Codex or Claude using my subscription.
What’s running Checking the current run…
Your run is live Watch your robot here. The video may follow with a short delay. —:— Stop run
⛶
Top Connecting
⛶
Observer Connecting
⛶
Left Connecting
⛶
Right Connecting
Connecting
—:—
Waiting for a run
Run
Waiting for a run
Astra’s next note will appear here.
Task chat
Visitor chat Close
Chat with other people watching. Messages are public.
No messages yet. Say hello!
Connecting…
Recent messages clear when the server restarts.
Runner: —
Prompt Waiting for someone to run a policy.
Astra’s replies and robot feedback from the current run appear here for everyone.
Your task
+ Connecting… Stop ↑
Select model and run Run via own LLM subscription
Run online Run Locally through Subscription Run with Astra Run with Opus Run GR00T
You can intervene here and update the prompt on the go. Updates guide the next decision. Motion already sent may finish; use Stop to halt.
Settings
Model execution
Automatic · BLUPE runner BLUPE runner Robot controller
The existing BLUPE runner collects observations and calls the model.
Controller model or protocol
Provider or policy
ChatGPT Claude No AI — just wiggle arms Astra · bring your key Custom end-point
Thinking effort
Low — fastest Medium — balanced High — more reasoning Extra high Max Ultra
Applies to your next run. Higher effort may improve decisions but takes longer.
Calling mode
Sparse calling
Normal calling
Sparse calling keeps the newest 1st, 2nd, 3rd, 5th, 7th and 10th completed calls. Normal calling keeps the full run history.
Inference speed
Fast Standard
Fast uses the provider's faster inference tier where available. It can use more credits; thinking effort and arm speed are separate.
System prompt ↗
Starting prompt
Instructions for this model
Saved for this model in this browser. Applies to your next run. Robot limits still apply.
View reference run
Reset to default Cancel Save changes
Use an API key instead
Continue with ChatGPT You have used your free guest runs. Use your API key or run locally to continue.
Sign in with ChatGPT and run Keep watching
Check Codex connection First-time sign-in: ./run.sh --codex-login --provider codex
Check Claude connection First-time sign-in: ./run.sh --claude-login --provider claude
Joint motion cap 20° / 100 ms
Larger moves are slowed with extra waypoints, preserving the path and final target.
The model observes the cameras, chooses a move, and waits for robot feedback before deciding again.
Email (optional)
This project is super new. We'd love to follow up about your experience.
Share conversation on the public playground so others can see
Paste the following prompt into your favorite coding agent:
Setup prompt
Copy setup prompt Check again
Checking the connection uses a small amount of subscription usage.
Forget key & stop Call operator
Connecting to your browser session…
Prompt
Runner: —
Waiting for someone to run a policy.
Astra’s replies and robot feedback from the current run appear here for everyone.
Conversation Download log Watch replay
Replay becomes available after the robot uploads its video.
Conversation with Astra
Recorded prompts, model replies, and robot feedback for this run. Download the log for full request and response data.
No model conversation recorded yet.
Your model decisions and robot feedback will appear here.
Runs from your current session.
Your runs will appear here.
More of your runs
Models: All models
Choose one or more models
All models
Select All models to clear the model filter. On desktop, hold Command or Ctrl to select several.
Search videos by task
Search Clear search
Find recordings by their task prompt, such as apple or put pens into a cup. Related objects and partial tasks may match. Watch a replay to see what happened.
Loading past runs…
Hover for a quick preview. Select a recording to watch the full run.
Nearby points have similar task descriptions; this does not establish identical starting conditions or physical difficulty. Small offsets separate repeated attempts.
Reset view Zoom in Zoom out Green: Success Red: Failure Hollow: No agreed result
Hover or focus a run for a muted video preview; click or press Enter for full replay. Drag to rotate; pinch or scroll to zoom. Arrow keys rotate, +/− zoom, Home resets. Use Previous/Next or [ and ] to focus each run; Escape closes the preview. Hollow points have no agreed result. Table provides an accessible alternative.
Previous run Next run Open focused run
Select a point to inspect a run.
Load more
How to use the playground
Choose a robot and model. Open the model menu to set up your connection and run duration.Describe the task. Send your prompt to join the queue. Watch the cameras and follow the robot’s progress in task chat.Review the run. Browse recordings below, or open the dataset to explore the camera and action data.
API docs ↗ Run locally ↗ ASPIRE README ↗