Run Tasks with MoveIt Pro Orchestrator (Beta)
MoveIt Pro Orchestrator is in beta and subject to change: its tool, program format and results may change in a later release.
MoveIt Pro Orchestrator lets an AI assistant connected through the MCP server carry out tasks on your robot. It is off by default.
How It Works
The LLM works in a cycle: sense, plan, act, and measure. Orchestrator compiles its agentic programs into MoveIt Pro Objectives.
It needs no per-robot setup: it reads what your robot can do from the robot itself.
What It Can Do
- Move one or several arms: planned motions, straight lines, and relative moves.
- Pick and place objects, planning around what depth cameras see on robots that have them.
- Push and turn until a set force or torque on robots with a force sensor, and pull a drawer or door open by its handle.
- Navigate a wheeled base with Nav2 on robots that run it.
- Run your existing Objectives, including learned policies, and any single Behavior.
- Report where the tool and named objects ended, whether an object tipped over, and camera images.
Enable It
Add --orchestrate when you install or start a local or SSH MCP session, then reconnect your client:
moveit_pro mcp install --client claude-desktop --orchestrate
The session offers orchestrate and save_objective for requested program saves. For the best results finding objects, also enable Gemini scene inspection with a key from Google AI Studio. Without the flag, a session still runs existing Objectives with run_objective. With it, the session leaves out general authoring and host lifecycle tools, get_capabilities and run_objective, since an objective step runs an Objective. The read-only HTTP connection cannot use it. A session restricted by a policy file lists runtime:execute and runtime:orchestrate instead of using the flag.
Orchestrator lets the assistant command robot motion and run any Behavior. Your MCP client's tool approval is the only check on each call. Try a task in the MoveIt Pro Simulator first, and keep Stop Motion within reach.