Execution and cleanup#
Warning
If hardware is moving unexpectedly, use the workcell’s physical emergency stop and established safety procedure. Browser controls, ROS messages, and process termination are not a safety-rated stop.
Another GTA server is detected#
The supervisor checks local processes and the ROS graph and refuses to take
over another GTA server. Identify the owner and stop it deliberately before
using Run. Do not launch scene/robot/server.py separately for a managed run.
External camera and gripper drivers should remain running.
Both dashboards in the same checkout share an execution lock. Separate checkouts and personal hardware scripts are not made safe by that lock. Do not delete a lock file as a substitute for checking active processes.
ROS probe fails before the server starts#
An import/executor/context traceback can mean incompatible Python/ROS libraries. Verify the application’s interpreter, sourced Humble environment, and robot workspace using ROS 2 setup.
Publishers time out or the server never becomes ready#
All five robot command subscriptions must match. Verify GTA_BACKEND,
domain ID, the selected robot implementation, and server logs.
“Subscribed” is only one startup step: gripper action availability, RTDE
control, and fresh robot state must also succeed.
The managed supervisor waits up to 30 seconds. The gripper wait defaults to 10 seconds and client subscription matching to 5 seconds. Inspect the first failed startup step using the endpoint checks.
Open3D cannot open the preview#
A local display and functioning OpenGL context are required. Headless shells, remote display forwarding, and mixed graphical libraries can produce GLFW/GLX errors. Test display/library configuration outside a robot run. A preview startup failure stops the script. Cancel remains available while the preview is open.
Skip does not work during preparation#
Skip applies only to an active subtask. It is disabled while the server starts, the script prepares perception, or the 3D preview is open. One pending Skip is allowed, tagged to the run and subtask, so repeated clicks cannot skip a later subtask. Cancel stops the entire run instead.
“Exited without confirming hardware cleanup”#
This is a failure, not successful completion. Inspect the server log and structured execution events for the first error. Cleanup must acknowledge stopping force/servo control, cancelling goals, and releasing owned resources. A crash, missing cleanup event, failed stop, or forced termination can prevent confirmation.
The run remains Stopping until its processes exit. After the five-second graceful window the supervisor may forcibly terminate owned processes, which does not prove a hardware stop succeeded. Check the robot/controller state using the workcell’s safety procedure before trying again. Do not suppress the message or start another server while cleanup is unresolved.
Logs are retained in the workspace’s execution_logs/<run-id>/. Review
storage and privacy before sharing them.