Program for calculating the linear drag coefficient in all 6 directions #394

JackGonzo11 · 2019-04-27T02:32:48Z

No description provided.

Program to calculate the linear drag in all 6 directions.

kev-the-dev

Great work on this program! After some small changes I hope to see it running on the Sub to model this and help to improve our controls!

In addition to my comments, the program should go in gnc/sub8_system_id/nodes/estimate_drag_coefficients and info should be added to the README of that package explaining what the program does and how to use it.

kev-the-dev · 2019-04-27T04:47:59Z

Drag.py

+rospy.set_param('LINEAR_FORCE', 10)
+rospy.set_param("TORQUE", 5)
+rospy.set_param('ForceDown', -20)
+rospy.set_param('TimeOfForceDown', 10)


Instead of setting them here, just get them as you do below, with defaults. This code would prevent them from ever being changed, as they're just getting reset when the program runs

It may also be wise to consider moving this to a config down the line, and ensuring sensible values are set before starting anything.

kev-the-dev · 2019-04-27T04:49:44Z

Drag.py

+
+
+def Find_Bouyancy(choice):  # Initial function used to upward force of buoyancy (used in determining drag in Z axis)
+    global Bouyancy


Not great practice to make all your variables global, this would definitely be much more readable using a class

kev-the-dev · 2019-04-27T04:52:07Z

Drag.py

+        Apply_Force('yw')
+        Apply_Force('rl')
+        Apply_Force('p')
+        # Drag coefficients are written to file, do with them what you wish.


I noticed there is no user feedback for this program. It would be kinda scary/confusing to watch the robot do this with no idea why it's moving or what it's thinking. Encourage you to use rosypy.loginfo upon making these transitions of state. Also recommend printing out to the user the solution in addition to writing it to a file

kev-the-dev · 2019-04-27T04:55:19Z

Drag.py

+        file_object.write("\nRoll: " + str(Roll_Drag))
+        file_object.write("\nPitch: " + str(Pitch_Drag))
+        file_object.write("\nYaw: " + str(Yaw_Drag))
+        file_object.close()


Would be more helpful if this was written to a file in a way that a computer could read, such as yaml. Luckily, this is super easy in python!

import yaml data = {'x': 500, 'y': 500, ...} f = open('drag_coefficients.yaml', 'w') yaml.dump(data, f)

Ditto'ing Kevin's comment, definitely consider dumping the data into a format that's computer readable, a yaml like what Kevin mentions would be great. It would make it much easier to dump things into Python to extract data, graph, etc.

kev-the-dev · 2019-04-27T04:56:26Z

Drag.py

+        file_object.write("\nYaw: " + str(Yaw_Drag))
+        file_object.close()
+    except rospy.ROSInterruptException:
+        pass


Passing on exceptions is a major red flag. The user would not be notified of why the program closes or potentially unpredictable behavior of the robot that an exception could cause. You should always atleast

except rospy.ROSInterruptException as e: rospy.logerr(e)

kev-the-dev · 2019-04-27T04:57:18Z

Drag.py

+    Calculates drag based on the initial applied force and the approximate max velocity
+    the sub achieves in that axis. See for formula:
+    Axis determined by char argument.
+    '''


Appreciate your use of docstrings. It also really nice to talk about what the functions parameters are and what, if anything the function returns.

Also it seems like the formula is missing

jpanikulam

This is exciting! Jason and I talked about doing something like this for a long time.

Some things to consider:

Observe many, repeated motions of the sub and try to fit a model. Read up on optimization, your keywords are "maximum likelihood estimation" and "system identification"
The measurements you get from the real sub will be noisy, exact < 0.0 checks that work in no-sensor sim won't work in the pool

jpanikulam · 2019-04-27T05:05:07Z

Drag.py

+rospy.set_param("TORQUE", 5)
+rospy.set_param('ForceDown', -20)
+rospy.set_param('TimeOfForceDown', 10)
+Linear_Drag_X = 0


Consider naming things in a consistent way

Variables: snake_case
Functions: snake_case
Classes: CamelCase

jpanikulam · 2019-04-27T05:08:34Z

Drag.py

+                pub.publish(force_msg)
+                Down_Force = Down_Force - .001
+                rospy.sleep(.01)
+                if Velocity < 0.0:


Remember that the sub's velocity is estimated from sensors, so it will be noisy. Even if the sub is perfectly still, you'll observe a velocity bouncing around zero.

One might expect this to be compounded by the fact bobbing of the sub in the water

jnez71 · 2019-04-27T06:28:20Z

As @jpanikulam mentioned, the right way to do this is to formulate it as a statistical regression problem. Not only would you get better results with less painstaking experiments, you'd also become an omnipotent god of estimation theory in the process.

However, MIL is 50% about doing things right, and 50% about doing things right enough. So in the style of your approach here, might I suggest the following:

It sounds like the real goal is to get a simplified drag model that can help compute a feedforward term in your controller. I.e. your PD controller will output some wrench based on feedback, plus the drag coefficients you found here times the current twist, in hopes of cancelling the effect of drag on the sub.

So, just skip ahead and add that term to your controller right now with a wild guess at the drag coefficients (all zero is a fine guess). Then, instead of commanding wrenches for experimenting (which wont stop the sub from moving along axes it shouldn't for the current test), you should command twists through your controller. So if you're testing the sub's forward motion, command a twist that is only nonzero for forward linear velocity.

As usual, you'll see a nice controlled straight line motion that rejects most bobbing and other disturbances... but also as usual, the velocity that the sub achieves won't be the exact velocity you commanded! It will be off by some amount, because your feedforward term has the wrong drag coefficient. Tweak the drag coefficient until the sub's velocity matches what you commanded. Repeat for each axis.

If tweaking by hand seems tedious, code up your tweaking process so it happens automatically. Congratulations, you made an adaptive controller.

Anyway, nice work so far!

rleyva · 2019-04-27T08:39:34Z

Drag.py

+Max_Velocity = .1
+Bouyancy = 0
+# Magnitude of applied force for determining drag coeffcients in linear axes
+appliedLinearForce = rospy.get_param("LINEAR_FORCE", default=10)


I'd caution against setting defaults that would cause the sub to do something when no config/values are set (in this case I'm assuming there would be some force/torque/etc. applied for some amount of time?)

tessbianchi

Great work! Happy you want to join MIL! Send your resume our way if you are looking for an internship!

tessbianchi · 2019-04-27T17:10:06Z

Drag.py

+    Calculates drag based on the initial applied force and the approximate max velocity
+    the sub achieves in that axis. See for formula:
+    Axis determined by char argument.
+    '''


Also it seems like the formula is missing

tessbianchi · 2019-04-27T17:12:00Z

Drag.py

+    if (choice == 'x'):
+        Linear_Drag_X = (appliedLinearForce / abs(Max_Velocity))
+    elif (choice == 'y'):
+        Linear_Drag_Y = (appliedLinearForce / abs(Max_Velocity))


I don't see a from __future__ import division at the top of the file. This can lead to some annoying bugs in the future if an int is passed into this function. If you still want to keep the int dividing behavior in other parts of the file, cast one of the values to a float before dividing.

tessbianchi · 2019-04-27T17:12:13Z

Drag.py

+
+
+def Find_Bouyancy(choice):  # Initial function used to upward force of buoyancy (used in determining drag in Z axis)
+    global Bouyancy


…ms/id/estimate_drag_coefficients.py

…py to gnc/sub8_systems_id/sub8_systems_id/estimate_drag_coefficients.py

….py to gnc/sub8_system_id/sub8_system_id/estimate_drag_coefficients.py

JackGonzo11 · 2019-05-08T14:35:06Z

I added the new program along with a readme file to my forked repo. Let me know if I need to change anything before it gets merged!

kev-the-dev

Nice improvements to readability! It does not seem to run for me, gets past the buoyancy stage then gets stuck on applying a Z force (nothing is published to /wrench).

I'd recommend running this on your machine and see if you can reproduce this problem. It may be related to you recreating Pub/Subs within a loop or init_node'ing twice

gnc/sub8_system_id/ReadMe.md

kev-the-dev · 2019-05-09T00:07:18Z

gnc/sub8_system_id/sub8_system_id/estimate_drag_coefficients.py

+        terminal (linear/rotationl) velocity in that direction. Once that
+        max velocity is found, it is used in the calculate_drag() function.
+        '''
+        global max_velocity, bouyancy, linear_drag_z


Still using the globals here but they should be a part of self

kev-the-dev · 2019-05-09T00:09:47Z

gnc/sub8_system_id/sub8_system_id/estimate_drag_coefficients.py

+    def find_bouyancy(self, choice):  # Initial function used upward force of buoyancy(used determining drag in Z axis)
+        down_force = -.5  # Applies initial downward force in z axis
+        pub = rospy.Publisher('/wrench', WrenchStamped, queue_size=30)
+        rospy.init_node('move_sub', anonymous=False)


init_node must only be called once at the start of the program before and sub/pubs are created

kev-the-dev · 2019-05-09T00:10:22Z

gnc/sub8_system_id/sub8_system_id/estimate_drag_coefficients.py

+        force_msg.wrench.force.z = 0  # Applies 0 force which stops downward force
+        pub.publish(force_msg)
+        while not (rospy.is_shutdown()):
+            rospy.Subscriber("/odom", Odometry, self.get_velocity, choice)


Since you have these in multiple parts of your program, perhaps they should be a class variable

JackGonzo11 · 2019-05-10T23:20:40Z

I made the ReadMe file correction. I also addressed the problem about nothing being published to wrench. Apparently a newly declared publisher has to wait a small amount of time so the topic can connect before it sends a message

kev-the-dev · 2019-05-11T22:31:06Z

Most of the problems in my latest review still seem present.

The program gets stuck on "Applying a force in the z direction"
Still suscribers created within loops which will likely break things
Program still does not have execute permissions. I highly recommend using the git command line rather than github's file upload for things like this. When you change a file's permissions, you have to commit this change just like any other change

JackGonzo11 added 2 commits April 18, 2019 16:00

Add files via upload

d9813fb

Program to calculate the linear drag in all 6 directions.

Add files via upload

e642a66

kev-the-dev requested changes Apr 27, 2019

View reviewed changes

jpanikulam reviewed Apr 27, 2019

View reviewed changes

rleyva reviewed Apr 27, 2019

View reviewed changes

tessbianchi reviewed Apr 27, 2019

View reviewed changes

JackGonzo11 added 6 commits May 8, 2019 10:24

Delete Drag.py

0a3413d

Add files via upload

649ff25

Rename ReadMe.md to gnc/sub8_system_id/ReadMe.md

2a9e485

Rename estimate_drag_coefficients.py to gnc/sub8_sytems_id/sub8_syste…

c22eb2c

…ms/id/estimate_drag_coefficients.py

Rename gnc/sub8_sytems_id/sub8_systems/id/estimate_drag_coefficients.…

07dda18

…py to gnc/sub8_systems_id/sub8_systems_id/estimate_drag_coefficients.py

Rename gnc/sub8_systems_id/sub8_systems_id/estimate_drag_coefficients…

28dad8d

….py to gnc/sub8_system_id/sub8_system_id/estimate_drag_coefficients.py

kev-the-dev requested changes May 9, 2019

View reviewed changes

JackGonzo11 added 7 commits May 10, 2019 15:46

Add files via upload

4484708

Delete ReadMe.md

953c1cd

Delete estimate_drag_coefficients.py

ef80a26

Add files via upload

f8632bd

Add files via upload

45faf6c

Add files via upload

d7db1e2

Add files via upload

7cb6125

JackGonzo11 added 4 commits June 6, 2019 14:49

Merge branch 'master' into master

1c27115

Add files via upload

3ce43d6

Delete estimate_drag_coefficients.py

9aa024b

Update estimate_drag_coefficients.py

7f775f7

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Program for calculating the linear drag coefficient in all 6 directions #394

Program for calculating the linear drag coefficient in all 6 directions #394

JackGonzo11 commented Apr 27, 2019

kev-the-dev left a comment

kev-the-dev Apr 27, 2019

rleyva Apr 27, 2019

kev-the-dev Apr 27, 2019

tessbianchi Apr 27, 2019

kev-the-dev Apr 27, 2019

kev-the-dev Apr 27, 2019

rleyva Apr 27, 2019

kev-the-dev Apr 27, 2019

kev-the-dev Apr 27, 2019

tessbianchi Apr 27, 2019

jpanikulam left a comment

jpanikulam Apr 27, 2019

jpanikulam Apr 27, 2019

jnez71 commented Apr 27, 2019

rleyva Apr 27, 2019 •

edited

tessbianchi left a comment

tessbianchi Apr 27, 2019

tessbianchi Apr 27, 2019

tessbianchi Apr 27, 2019

JackGonzo11 commented May 8, 2019

kev-the-dev left a comment •

edited

kev-the-dev May 9, 2019

kev-the-dev May 9, 2019

kev-the-dev May 9, 2019

JackGonzo11 commented May 10, 2019

kev-the-dev commented May 11, 2019



		def Find_Bouyancy(choice): # Initial function used to upward force of buoyancy (used in determining drag in Z axis)
		global Bouyancy

Program for calculating the linear drag coefficient in all 6 directions #394

Are you sure you want to change the base?

Program for calculating the linear drag coefficient in all 6 directions #394

Conversation

JackGonzo11 commented Apr 27, 2019

kev-the-dev left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jpanikulam left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jnez71 commented Apr 27, 2019

rleyva Apr 27, 2019 • edited

Choose a reason for hiding this comment

tessbianchi left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JackGonzo11 commented May 8, 2019

kev-the-dev left a comment • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JackGonzo11 commented May 10, 2019

kev-the-dev commented May 11, 2019

rleyva Apr 27, 2019 •

edited

kev-the-dev left a comment •

edited