History

Old

Ein

2: Shift to ready position on the front table. capslock + @: assume wholeFoodsCounter1 c r c: clear call stack capslock + shift d: queues scan program y: execute what is in the queue capslock + page up/page down: select target to pick u: display where the end effector is located capslock + q: vision cycle capslock + b: toggle publish objects capslock + f: reinitialize and retrain everything capslock + n: listen for pick requests from fetch command capslock + v: thompson sampling test capslock + I: count grasp

capslock + w: queues pick program

capslock + -: samples from Thompson capslock + =: empty-a

numlock + X: Organize windows capslock + z: load target class range map into register 1 numlock + 6: select best available grasp (not used) numlock + ,: select best available grasp (used by capslock + w) capslock + t: synchronic servo capslock + Y: quick fetch taragetClass numlock + a: drawMapRegisters numlock + +: shift grasp gear

Grasp gear 2 on handle is a good grasp.

Saving candidates: capslock + A: capslock + numlock + u: capslock + numlock + A: SAVE GRASP ME capslock + Z capslock + numlock + i: set grasp memories from classGraspMemories

numlock + s: select max target NOT cumulative numlock + S: select max target cumulative S capslock + numlock + s, d, f: change current pick mode

capslock + numlock + ;: height learning

Command:
catkin_make && gdb --args ./devel/lib/node/ein _data_directory:="$(rospack find node)/ein_data1" _vocab_file:="vocab.yml" _knn_file:="knn.yml" _label_file:="labels.yml" _left_or_right_arm:="left" _gray_box_left:=90 _gray_box_right:=90 _gray_box_bot:=60 _add_blinders:=0 left

General notes:

crc: clears the stack y: Runs what is on the stack

Training new model: capslock + shift d: queues scan program

Prior (grasping): 2: move to position 2 capslock + pageup: select target class capslock + backspace: load prior capslock + numlock + s: prior capslock + w: queue pick command

Training (grasping): 2: move to position 2 capslock + pageup: select target class capslock + backspace: load prior (if desired) capslock + numlock + d: select learning sampling capslock + w: queue pick command capslock + numlock + A: save the learned numbers

Marginal (grasping): restart the program 2: move to position 2 capslock + pageup: select target class capslock + =: load marginals. Verify they are in Grasp Memory Sample window. capslock + numlock + d: select learning sampling capslock + w: queue pick command

Training height: capslock + pageup: select target class capslock + numlock + backspace: load prior (if desired) capslock + numlock + :: queue height learning

---

ein_data_height_done (height, no grasp, possibly bad contrast) ein_data_contrast (no height, no grasp, good contrast) ein_data (baseline scans, possibly bad contrast) ein_data_combine1 (height, no grasp, fixes)

• really good models

Ein MIT Visit

Pull. Copy the folder node/default/objects to node/ein_dataDefault. Copy the contents of node/default/config to node/ein_dataDefault.

From the root of your catkin workspace, run the following command:

catkin_make && gdb --args devel/lib/node/ein _data_directory:="$(rospack find node)/ein_dataDefault" _vocab_file:="vocab.yml" _knn_file:="knn.yml" _label_file:="labels.yml" _run_prefix:="ISRR" _left_or_right_arm:="left" left

And enter 'r' to start the program from within gdb. If the robot is enabled it should move to a ready position.

In a different window run:

rosrun node ein_client.py left

to start the repl. Enter:

fillClearanceMap loadIkMap ;

and the Object Map View should change to reflect a mapping of a cross section of the ik space. There should be red, light red, yellow, and green dots. Note, a space between the last word and the semi-colon is necessary.

Make sure numlock and capslock are off. With the Object Map View focused, use 'a','d','q', and 'e' to move the arm deep within green territory. Place an object directly under the arm. Run the following command in the repl:

scanObject ;

and follow the instructions in the ein terminal window to scan the object.

At any time, run

guiShowAll ;

from the repl to show all of the windows.

When scanning finishes, models will be trained with an object class corresponding to each subfolder of ein_dataDefault. Since there are no RGB images for background, blueBowl, brownCup, and brush, models will be trained for those classes but with no positive examples for them, nothing will ever be classified as one of those objects. So those folders can be removed at this point (they were necessary earlier for picking to work with those classes and the pretrained knn and vocab).

loop example:

 0 10 start waitUntilAtCurrentPosition xDown xDown xDown xDown waitUntilAtCurrentPosition xUp xUp xUp xUp next ;