Referenced builds | https://drive.google.com/drive/folders/1xHWeQ9WROf7ZX0CzsEsIwfUtTN2c4YTj |
|---|---|
Date |
|
Point of contact |
Known feature limitations
Noise can have an impact on robot motion. Lighting issues, reflective materials etc. may falsely appear as obstacles. Robot moves may be 2x slower as compared to RapidPlan robot without RapidSense.
Although the system theoretically can support an unbounded number of cameras and robots, we have been focusing on supporting systems with 4 cameras and 2 robots.
RapidSense does not provide guidance on number of cameras, camera placement, or lighting.
RapidPlan and RapidSense will be limited by the accuracy of the “digital twin”. Misalignment between the real and virtual world may lead to slower cycle times due to recalculating paths to avoid parts of the scene that aren’t represented in the model but “identified as obstacles” through RapidSense. Creating/maintaining an accurate digital representation is not part of the RTR scope.
Extend the RapidSense robot filter to filter tools/attachments to the robot has to be configured manually and only boxes can be used.
This system setup is not yet available in a user-friendly UI, but clear documentation has been created on the process. RTR support may be required for initial RapidSense setup.
RapidSense tools do not provide guidance on if calibration targets are effective for calibration.
This is especially important given a known bug which will falsely report a successful calibration (details below)RapidSense does not have ability to recognize expected vs. unexpected objects aside from robots, so setup will have to disable collision checking on the pick motion. That means if the pick location is blocked, or the part is in the wrong orientation, the robot will collide with it.
If something is resting on the robot, it will be filtered out and the robot will continue moving.
RapidSense is not safety certified for collaboration with humans
RapidSense is only supported for GPUs with double-point precision (nvidia anything GTX 4xx forward, AMD anything beyond Radeon Pro VII, most GPUs from 2020 or later should not have any issues)
Known bugs
Calibration will show as succeeded even if it couldn't get a valid image of the cal plate. Even if the plate is in view, the camera can be too far from the plate. To avoid this, a user should use other tools such as the Frame Viewer to validate that their calibration targets are visible/valid.
If a camera is unplugged, RapidSense may crash or keep running, it's inconsistent.
Robots running at high speeds may see the robot stop unexpectedly due to latency in the robot filtering. Robots up to 1.5 m/s tool speed should not see any issues.
There are some cases where the robot filter does not start running. This has not been seen on the linked build but just in case the current workaround for this is to re-start the proxy service.
In some cases, voxels may remain outside of a volume when editing volume bounds.
If starting the calibration service while already at the first calibration target, the calibration service misses the delayed response. Recommended suggestion to avoid this is to start calibration from home instead.
Calibration will attempt to calibrate a camera that is no longer connected and then hang
Resolved in newer RS versions
Changes which minimize RS crashes on start
Planned improvements
None yet