Robotics

Gemini Robotics models produce action plans for physical or simulated robot agents. Sessions are opened at /v1/robotics/sessions.

Example

curl https://llmtr.com/v1/robotics/sessions \
  -H "Authorization: Bearer sk_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-robotics-er-1.5-preview",
    "input": {
      "goal": "Pick up the red block and place it in the tray."
    },
    "max_budget_usd": 1.00
  }'

Response

{
  "id": "robsess_xxx",
  "status": "succeeded",
  "actions": [
    { "type": "move_to", "target": [0.2, 0.3, 0.1] },
    { "type": "grip" },
    { "type": "move_to", "target": [0.5, 0.5, 0.2] },
    { "type": "release" }
  ]
}

Input schema

The input object accepts:

• goal — natural-language goal
• scene — optional scene description or visual file_id
• constraints — safety or motion constraints

Pricing

Billed by the native unit shown on the model card (usually tokens + action count).