X Axis Flip Still Needed?

[mdicicco]

Is it still true that ATI sensors don't use the right hand rule for their sensor axes? I have a new Axia80 and it seems to be correct (i.e. it uses a right hand coord system). The raw netft_data seems to match up with the +X and +Y labels on the sensor and the +Z data lines up with a right hand rule. The flip on lines 244/247 of netft_utils seems to be unnecessary.

I did a quick scan of the ATI website and first two sensor drawings in their pdf (the nano17 and then the nano17e) both use a right hand rule as well, although I don't have any other sensors to test on.

I modified my local copy but it could be a good thing to either note or put in a config to disable.

[jbeck28]

I reached out to an applications engineer with ATI, she told me that all of their sensors follow the right hand rule. So I would think the flip is unnecessary.

[riv-tnoble]

@AndyZe

We've fallen into the same pitfall due to the x axis inversion, and are having to use our own fork of the repo which does not apply the inversion.

It'd make sense to fix the issue at the source by either removing the inversion (if all are right-handed) or by providing the option to the user (if there's a mix of left-handed and right-handed).

I'm happy to raise a PR, but would appreciate input on which idea is better suited as I'm not familiar with the full range of force sensors ABI offers.

[mdicicco]

Just revisited this ticket for fun now many years later and I see that the driver has been updated for ROS2 (hooray!).
That means the X flip in question has now moved lines and is now done here:
https://github.com/UTNuclearRoboticsPublic/netft_utils/blob/ddccade65049195324c384919be7391b24929b48/netft_utils/src/netft_utils.cpp#L197C1-L217C52

[zimmerman-pliant]

I finished setting this up in our ROS 2 system and I think that netft_utils.cpp and netft_utils.h may not actually be used at all anymore, and the axis flip above seems to occur in dead code.

I see two ways to use the package:

1. There's ros2_control integration which doesn't apply any flip here in hardware_interface::return_type NetFTHardwareInterface::read(...):

netft_utils/netft_utils/src/netft_hardware_interface.cpp

Line 92 in ddccade

hw_sensor_states_.at(0) = wrench.wrench.force.x;

2. And there's a standalone node that can be run with ros2 run which uses the split-out netft_rdt_driver C++ package:

netft_utils/netft_utils/src/netft_node.cpp

Line 41 in ddccade

#include <netft_utils/netft_rdt_driver.h>

I think the netft_rdt_driver is getting the RDT data to the member variable new_data_ here, again without an axis flip:

netft_utils/netft_utils/src/netft_rdt_driver.cpp

Lines 456 to 464 in ddccade

	tmp_data.wrench.force.x = static_cast<double>(rdt_record.fx_) * force_scale_;
	tmp_data.wrench.force.y = static_cast<double>(rdt_record.fy_) * force_scale_;
	tmp_data.wrench.force.z = static_cast<double>(rdt_record.fz_) * force_scale_;
	tmp_data.wrench.torque.x = static_cast<double>(rdt_record.tx_) * torque_scale_;
	tmp_data.wrench.torque.y = static_cast<double>(rdt_record.ty_) * torque_scale_;
	tmp_data.wrench.torque.z = static_cast<double>(rdt_record.tz_) * torque_scale_;
	{
	boost::unique_lock<boost::mutex> lock(mutex_);
	new_data_ = tmp_data;

The node calls netft-getData(data) here:

netft_utils/netft_utils/src/netft_node.cpp

Line 124 in ddccade

netft->getData(data);

That just writes new_data_ to the passed-in data:

https://github.com/UTNuclearRoboticsPublic/netft_utils/blob/ros2/netft_utils/src/netft_rdt_driver.cpp#L481-L487

Empirically the package runs as-is and publishes the correct wrench with an Omega85 sensor, which I can confirm is definitely using a right-handed coordinate system.

I also deleted netft_utils.cpp and netft_utils.h as a test. The package builds and compiles and both the ros2_control integration and the standalone node run fine and correctly report the forces and torques consistent with the sensor frame in the Omega85 Docs (ati-ia.com).