paper: FRC #33 The Killer Bees 2013 Software - BuzzXVIII (Buzz18)

[apalrd]

Thread created automatically to discuss a document in CD-Media.

FRC #33 The Killer Bees 2013 Software - BuzzXVIII (Buzz18) by apalrd

[apalrd]

As promised, a complete release of Buzz18 software.

I am quite busy with school at this time, but I should finish the promised papers within the next week.

For now, the software zip is posted in CD Media.

[z_beeblebrox]

Not having LabVIEW, I can't look at the code. However, a while ago, you mentioned that you used something other than your normal "Cheesy Drive" software for drivetrain control. What did you do?

[apalrd]

Added file buzz_lib_sw_2013.zip (BuzzLib 2013)

New functionality from last year:
-Simple Vref Diagnostics - Used with shorting jumper on adc7 (between 5v and analog pin), used to diagnose simple faults of 5v rail and provide ratiometric multiplier for ratiometric sensors.
-Sharp IR - Scaling blocks for all three Sharp IR sensors sold by Sparkfun. Analog to cm values copied directly from datasheet and interpolated using multipoint linear interpolator (see below)
-2d interpolation - Interpolates a 2d curve from a prelookup and 1d array of Z points. Also variants to round down and round up to nearest Z point.
-3d interpolation - Interpolates a 3d map from two prelookups and a 2d array of Z points.
-Prelookup - Finds the index and fractional index from a x input and 1d array of X breakpoints. All breakpoints must be spaced least->greatest, there is no checking of this. Used for 2d and 3d interpolation, as it is the most computationally intense step a single prelookup can be shared by multiple interpolations who share the same axis breakpoints.
-Accel calculator - Calculates the acceleration of a velocity assuming 10ms iteration time. Really just finds current derivative, but used for acceleration of shooter wheel speed.
-Hysteresis - Threshold with hysteresis.


The 2d and 3d interpolation tables are the most interesting. For an example of their use, see the Sharp IR blocks in BuzzLib, as they do 2d interpolation in the most basic form.

[apalrd]

Originally Posted by z_beeblebrox

Not having LabVIEW, I can't look at the code. However, a while ago, you mentioned that you used something other than your normal "Cheesy Drive" software for drivetrain control. What did you do?

That would be the 'CulverDrive', which we have a paper and example C code in-progress for.

The basic idea is to use the angle of the steering stick instead of the X axis of the steering stick as the primary steering input, when combined with a joystick with circular boundaries (such as the Logitech F310) this allows you to 'steer' the joystick around the boundary of the stick with finer steering resolution.

[apalrd]

Added new files:
buzz18_pal_lib.pdf
pal_lib_default.zip (LV example code, pal_lib code, and paper pdf)

A little bit about pal_lib:
-'pal_lib' WPIlib replacement for LabVIEW
-Replaces ENTIRE WPIlib LabVIEW with 30 VIs, contained in zip folder and inserted as a subfolder into your project.
-Heavily optimized architecture promotes execution efficienty, runtime performance determinism, and good embedded programming practices. Increases in execution efficiency are primarily done by reducing code size and removing unnecessary code, which significantly decreases compile and download times as an added benefit.
-FAST execution (all code currently executes at 10ms, for small or heavily optimized programs 5ms is possible). We could have run Buzz18 at 7ms iteration time, but ran at 10ms with ~55% CPU usage to leave overhead for LV probes and debugging.
-Supports basic commonly-used inputs, outputs, and control data. Large efficiency and code space gains are made by removing functionality that is not commonly used.
-Season-tested on Buzz18 (including five competitions and Einstein) at 10ms iteration time without any bugs EVER attributed to the library after initial library validation
-Extremely low CPU usage (initial validation tests provided initial functionality to default tank drive program with 1/2 CPU usage % at over 2x execution speed), library CPU usage does not increase with used functionality as the library is already handling all of the data it is capable of (e.g. adding another motor or encoder does not require any expensive VI calls and CPU usage of scaling function is inlined and negligible).
-zip contains example 'default' program which maps several joysticks, buttons, and limit switches to motors, solenoids, and relays, and operates the compressor. See comments in the code for exact mapping and IO pinouts. Entire example code is 4 VI's in addition to pal_lib
-Fully synchronous architecture forces efficient data-passing design, as all data can be passed via connectors. Use of global-visible data structures in Buzz18 further optimizes data-passing as all data is always available to all subsystems, any subsystem can change data without consequence to anything else, and order of operations is guaranteed (simplifies triggers significantly). Data structure architecture also simplifies dashboard programming, as all of the data structures are simply sent to the dashboard using the binary string method ('old-style', not NetworkTables) and the dashboard automatically has access to all key data.
-See buzz18_pal_lib.pdf for complete write-up
-Also see buzz18_sw_postcmp for complete robot usage of library

[apalrd]

OK, finished the 'CulverDrive' paper and attached it with example code.

What is CulverDrive?
-Improvement on previous 'Cheesy' and 'Halo' style drive systems, which are already used by many top teams
-Utilizes Logitech F310's round joystick boundary to provide a virtual steering wheel effect, allowing the driver to gracefully navigate the field
-Utilizes a mathematical algorithm similar to the 'Cheesy Drive', the intent in design was to replicate the functionality of a 'Cheesy Drive' using a real steering wheel on a joystick.
-Award-Winning (Waterford 'Innovation In controls' and Championship 'Quality' specifically noted CulverDrive in judge comments)

What does the zip include?
-LabVIEW source code, direct from Buzz18 software release, plus required BuzzLib interpolation blocks (prelookup and interp_2d)
-C source code, used in simulation and intended for Vex targets and readable without LabVIEW. Unfortunately it does not directly run on Vex because RobotC does support the C language very well.

Does it require a Logitech F310?
-In short, yes. The Logitech F310 has a circular joystick boundary, which is key to the algorithm. It's also smoother to drive than a square of octagon.
-The Xbox 360 joystick has an octagonal boundary, which will not provide the desired effect nearly as cleanly. The Logitech DualAction has a square boundary, which will not work very well at all.

As always, feel free to ask questions regarding any of our software.

[Iaquinto.Joe]

How do you open .lvsc files? Is it a tool included with Labview for FRC?

[apalrd]

LVSC file are LabVIEW Statecharts.

The module is not included in the FRC release of LabVIEW. We arranged this with NI in return for feedback on it's use.

Since I didn't expect anyone to have Statchart, I generated documentation for all Statecharts using the internal documentation tool. Every statechart has a folder called <name>_files which contains a PNG image of the following:
-The statechart itself (Main Diagram0.png)
-The Inputs and Outputs clusters (Inputs0.png and Outputs0.png) - I didn't use the StateData but there is a cluster for that too
-The outer appearance of each block (I tried to name them reasonably, that should be in these images - #.png)
-The entry/exit action of the block (for states - entryaction#.png and exitaction#.png
-The guard action of the block (for transitions - guard#.png)

Statechart was a great tool to work with once we learned all the quirks (e.g. when you copy-paste a block, they share the same diagram and modifying one modifies the other). It definitely helped some of the more complex state-machines we had (the first and second revision gun feeder system was very complicated). It was also helpful when debugging - I could tell the mechanical and integration people working on that system exactly what it was trying to do and how it should react.

I also made some VI's to graph the input/output data which was very helpful in analyzing the strange behavior (if I do this and this and this, it jams with gate A open and gate B closed but the gun isn't on...)... But even with the automation there were still corner cases that weren't handled perfectly. But the bucket (and really simple code) had virtually none of them.

[JamesTerm]

Originally Posted by apalrd

That would be the 'CulverDrive', which we have a paper and example C code in-progress for.

The basic idea is to use the angle of the steering stick instead of the X axis of the steering stick as the primary steering input, when combined with a joystick with circular boundaries (such as the Logitech F310) this allows you to 'steer' the joystick around the boundary of the stick with finer steering resolution.

That is brilliant! I am looking forward to reviewing this C example code!

[apalrd]

Originally Posted by JamesTerm

That is brilliant! I am looking forward to reviewing this C example code!

Example C code is in the zip (culverdrive_sw_release.zip). It isn't great, and isn't tested like the LV implementation, but should illustrate how the logic works.

interp.c and interp.h are C implementations of the LV interpolation functions. I tested these on a desktop GCC build/test harness and they work. We use these in the Culverdrive to define smooth enabling of different terms, and use them elsewhere for nonlinear functions. They're very useful functions. I can explain their use in more detail if anyone is interested.

culverdrive.h is the header, self explanetory.

The logic itself is written and commented in culverdrive.c in the culver_drive() function. The const data is defined in culverdrivedata.c.

There are some things that are a little weird about the C code. I intended to run it on Vex targets, but it won't compile in RobotC because RobotC does not support the C language very well.

[Iaquinto.Joe]

What would you recommend if a team wanted to implement a Finite State Machine approach to their programming? Should we try to talk with NI about the statechart plugin, or implement state machines on our own?

[JamesTerm]

Originally Posted by Iaquinto.Joe

What would you recommend if a team wanted to implement a Finite State Machine approach to their programming? Should we try to talk with NI about the statechart plugin, or implement state machines on our own?

Can you give an example of how you would like to use a finite state machine?

[Iaquinto.Joe]

Originally Posted by JamesTerm

Can you give an example of how you would like to use a finite state machine?

We drew out each subsystem's state machine so we would use a state machine for our shooter, our drivetrain, our pickup, etc. Having each of the systems diagrammed out before we program could help us debug and improve our code. We would have several independent state machines running in parallel (E.G. The drivetrain FSM runs in parallel to the shooter FSM).

[apalrd]

We use a lot of finite state machines for things that we sequence. We also use other control strategies for other things, including feedback controls, and lookup tables.

Our software modules are very large. They usually line up with the physical subsystem teams, although I threw the gun in with Roller because the integration between them was very critical and I wanted them to be designed together (there were 4 mechanical teams and 3 software modules).

We then design up a high-level design of the software (what each module is responsible for, what sensors and actuators it has command of, and what the interactions with other modules including HMI/Auton are). From that, we design each module.

The modules are split up into blocks, usually based on control of specific actuators or related groups of actuators. For example, drivetrain is split between drive motors and shift. Drive motors handles the drive motors, and shift handles the shifting. Drive motors is a passthrough function (control is unique to HMI/Auton, this is the only feature that is not implemented in it's module) and shifting is a lookup table. With servos in 2012, shifting was a finite-state machine that would peak and hold the servos for better shift quality and servo reliability (we melted servo or two before doing that).

So it dosen't make sense to design everything as an FSM. We really use a lot of small FSMs when we need to sequence actions at a high level (like the gun feed state machine or 2011 arm state machine), or deal with intermediate states at a medium to low level (like the peak-and-hold state machine).

All of the code runs in parallel (it all shares a single task, the software design strictly prevents blocking execution). The execution order is specified by how data is used by other modules.

I am not sure what NI's intentions are with the Statechart module.