Fix calibration script#193
Conversation
Signed-off-by: Tiffany Cappellari <tcappellari@theaiinstitute.com>
Signed-off-by: Tiffany Cappellari <tcappellari@theaiinstitute.com>
tcappellari-bdai
requested review from
heuristicus and
khughes-bdai
as code owners
Signed-off-by: Tiffany Cappellari <tcappellari@theaiinstitute.com>
Signed-off-by: Tiffany Cappellari <tcappellari@theaiinstitute.com>
mhidalgo-rai
left a comment
mhidalgo-rai
left a comment
There was a problem hiding this comment.
This a bit of a large patch and I have limited context. What is this fixing exactly?
| rgb_M_wr1 = cal["depth_to_hand"] @ hand_t_planning | ||
| # rgb_t_depth is depth_M_rgb (stereo extrinsic: maps RGB points → depth frame). | ||
| # depth_M_wr1 = depth_M_rgb @ rgb_M_wr1 | ||
| depth_M_wr1 = rgb_t_depth @ rgb_M_wr1 |
There was a problem hiding this comment.
@tcappellari-bdai meta: IIUC rgb_t_depth is in fact depth_M_rgb, so this isn't a bug. Still, even if the calibration dictionary uses a different convention, we should make sure ours aligns.
There was a problem hiding this comment.
agreed. its very confusing to use both types of notation.
There was a problem hiding this comment.
I had to use claude for help debugging and it seems to prefer the other convention 🥲
khughes-bdai
left a comment
khughes-bdai
left a comment
There was a problem hiding this comment.
I haven't reviewed thoroughly but I know you and Lu have been working on this for a while. If it's working on robot and @lgalup-bdai is good with the underlying math, I'll stamp it
|
folks, i cant visually verify if the math is correct. that would be like visually verifying that software has no bugs. is there something we can do to test it? |
|
looks like @tcappellari-bdai has run it on robot -- could you see an improvement with this vs the old method? |
|
@khughes-bdai Ya so previously, when i saved the new calibration params to the robot (after fixing the initial frame transform errors), the depth hand camera would just stop working. Like just sending empty messages and a black screen on the tablet. The calibration was also saying the depth and RBG cameras in the hand were like 5 cm apart (way too big). After these changes, the cameras are both fully functional and the frame positions and orientations look correct in rviz. (also Josh ran a spot check on Jasper after I changed the params and it passed all the checks, for whatever thats worth) |
4 participants
Fixes the calibration script (originally created by Gary) after a bug was discovered. Much of the fixes are based off of Lu's work in
bdai_calibration. Also updates the readmeTested on Jasper and Ishmael (their internal gripper camera params are set to the newly calculated ones from testing)