Troubleshooting

General

Q: The robot is in emergency stop, how can I reset it?

To reset the emergency stop:

Make sure all emergency stop buttons are released.
Press the white button on the control box.
In the GUI, go to the upper navigation bar and select: Emergency → Reset.

Q: How can I reboot the AI controller?

To reboot the AI controller, go to the upper navigation bar and select: PC → Reboot.

Q: When do I have to do a camera calibration?

The mounted cameras are calibrated during the commissioning process at NEURA. If you replace a camera or change its position, you must perform a new camera calibration to ensure accurate data and system performance.

Q: How can I check if the camera is working?

In the bottom-left corner of the navigation bar, you’ll find a camera button. Click it to display the current image from the connected camera. To diable, hit that button again.

Voice Commanding

Q: The voice commanding switch on the Sensors page is greyed out and I can’t enable voice commands. What should I do?

After starting or rebooting the robot, voice commanding requires some time to initialize. Please wait a moment until the system is fully ready—the switch will become active once voice commanding is available.

Q: The robot is not listening to my voice commands.

Make sure the robot is correctly activated by saying “Hey, MAiRA”. You should hear a beep—only start speaking after that sound.
Stand close enough to the robot to ensure your voice is clearly heard.
Always wait until the robot finishes speaking before giving your next command.

Q: How can I teach my own voice commands?

NEURA will soon provide a feature, to train the robot customized voice commands.

Workspace

Q: My taught workspace geometry doesn’t look like a proper rectangle. Why?: Ensure that you teach the points in the correct order and relative to the robot’s root link. The “left” point should be to the robot’s left when it is facing the positive X-axis, and the “right” point must be to the right of the left point from the robot’s perspective.
Q: Can I edit an existing workspace later?: Yes. In the Workspace Wizard, you can view and edit all existing workspaces.
Q: Can I create non-rectangular workspaces?: Currently, NEURA only supports rectangular-shaped workspaces.

Grasp Pose Recorder

Q: After click Detect object pose, application waits for long time

It may happen if you are trying to use object detecion or pose estimation model for the first time during the current session, as it takes time to set the model up. Please wait.

Q: Detect object pose failed

Please check if you loaded the correct object detecion and pose estimation model. Please also check if the object is fully visible.

Q: There is no object visualized in the GUI

Please check if <your object name>_downsampled.obj file exists.

Q: My hand pose not recorded correctly.

1. Please make sure you are using your pointing finger and thumb on your right hand. 2. Please make sure that these fingers are fully visible during the recording.

Q: My recorded pose is wrong in Robot mode.

Please make sure you are using the correct gripper and the offset of the TCP of the gripper is set correctly. Please make sure that ZeroG if turned off.

Q: I want to record grasps on the opposite side of the object.

You can save the current grasp. Then, you can select the Merge option to add new grasps, turn the object, and restart the recording process.

Q: My enrich failed.

It may happen if the origin recorded grasps are not close enough to a perfect grasping pose. Or the object shape is not suitable under the current parameter settings.

Try setting larger values for sample_points, touch_threshold, and leaf_size parameters, if recording graps for small objects.

Try increasing translation/rotation times for a better result (note that it will increase the computation time respectively).

Q: The pick is always too high (above the object) or just gently touched the surface of the object.

Check that:

1. max_aperture and default_opening are not too small. Sometimes due to the inprecise measurements, detected objects appear to be larger than real objects.

2. the workspace recorded is not too high for the object to be picked. Take a closer look at the point cloud visualized in GUI to ensure the object is fully shown: if not, reduce the height of the workspace.

A general fix is adding a small value to general_grasp_offset_z.

Q: The program quits right after I click Run button.

Some components may be started incorrectly, e.g., the spawn time of control was longer than expected, or the camera was not started at first.

Go to Supervisord page, which is the default page of the browser. Refresh the page and check if all components are running (especially motion planning and marker detection). If not, please restart them.

Q: There is a warning/error that the instance segmentation is not working.

The object detection model might be selected incorrectly. Quit the program, go to AI settings, and select the suitable model.

Q: There is a warning/error that no object is detected.

Check if the correct model for instance segmentation is selected. If so, please make sure the object is visible. If the problem persists, please retrain your object detection model with other lighting condition and background.

Q: The pick is not aligned with the object geometry.

Check if the object is close to the workspace boundary. If so, move the object to the middle of the workspace, or enlarge the workspace.

Note

If you change the workspace size, please make sure the updated workspace is fully visible from the current look-at point.

Please also check if the point cloud is of sufficient quality. E.g., if the object to be grasped is dark or shiny, under some lighting conditions the camera may work not as expected. In this case, please adjust the lightning to ensure the point cloud quality is reasonably good.

Q: There is always a warning about no collision-free pick.

Sometimes, possible picks can be mistaken for collisions due to point cloud quality. In this case, you could adjust default_opening (+-10mm).Alternatively, if using General picks, you could disable collision_pcd.

Q: Sometimes objects are picked successfully and sometimes not.

For different objects, the best results can be achieved with different grasp parameters. If your finger tip is flexible, you may also try disabling collision_workspace and collision_pcd.

Q: No valid grasp is generated even after long-lasting object detection.

Please first check if too many irrelevant points are visible. If so, adjust your workspace so that fewer such points are inside the workspace. For small objects, please make sure that the objects are still visible.

Bin Pick