Dog Swordfighting - Lesson 4: Strengthening the hold

So we ran into a problem.

Last time, Missile actually managed to get to a 15s hold pretty consistently. The issue was once we started transferring from the noodle to the toy sword...

Yeah he looked at me like I was an idiot and dropped it. Confused, I tried a few times, got a little dejected, and stopped.

So what's going on here?

As a general rule of thumb, if something isn't working it's because either:

1. The step was too large

2. The reward was too small

In other words, I'm either asking too much of him or paying peanuts.

I mean, maybe?

This is where some knowledge of the dog in front of me comes in to play. I'm not going to rule out that the treat isn't valuable enough for Missile. (He HAS been working his lemon brain after all)

Though Missile is also the kind of dog that accidentally spits out his food while training. I'm going to just assume that this is an understanding problem more than a motivation problem.

So where's the gap?

Let's think about it.

The object he's able to hold right now is a pool noodle.

• He can grab it anywhere

• It's fairly light

• It has an even weight distribution

The object I want him to hold is the toy sword

• He can only grab it in one location

• It's heavier than the pool noodle

• It has an uneven weight distribution

Just at a glance, the toy sword is completely different from the pool noodle, and Missile doesn't just give me a solid hold for fr-

You little shi-

As I was saying, Missile isn't one of those dogs that gives me a hold for free. (Usually) As of right now his default is to drop things.

Something unexpected? Drop.

I reach for something? Drop.

I blink too many times? Drop.

Suffice it to say if anything happens that he doesn't totally get, he will drop the toy.

This means that it boils down to preparing him for as many possibilities as I can. Enter generalization*

Chris WTF is "generalization"?

So hey! Cool, a new term.

For an official definition:

A scenario where the pet is able to perform the trained operant behavior in a variety of settings. Once a behavior is under stimulus control, trainers should endeavor to practise the behavior in different locations, with appropriate, gradually increasing, levels of 'distractors' (competing environmental stimuli). With respondent conditioning the conditioned response generalizes from the conditioned stimuli to other stimuli.

Source: "A Lexicon of Practical Terms for pet trainers & behavior consultants" Buy a copy over here, or don't, I guess.

Holy mother of- well, that was a lot of words.

The short version is that we want to test a skill against all sorts of possibilities to ensure it holds up. So for Missile that means we have to prepare him for the weird or confusing things that'll happen while he's holding the toy, ultimately making his hold significantly stronger.

Here's how we're going to do it:

How we're going to do it

He's able to hold the pool noodle. That's great, but he's just able to hold the pool noodle. It won't make the weebs or his ancestors proud.

First we can test it against various shifts he would naturally run into.

For the 3d nerds you probably know where I'm going with this, but for the rest of us, I'm going to:

• Test against a nudge on the noodle in the x, y, and z directions

• Test against a light rotation applied in the x, y, and z axes

Imagine this is Missile, just with fewer polygons.

The goal here is to mimic what's going to happen with the sword as he holds it, and ensure he doesn't just drop it the second something unexpected happens.

THEN we can add more weight.

• Add a light insert to the pool noodle

• Work back up to 15s

• Apply earlier nudges and rotations to noodle

THEN start adding more precision

• Add insert to shorter pool noodle

• Practice grabbing in one position

• Apply nudges and rotations in all directions

THEN start adding imbalance

• Shift insert to one side

• Continue practicing duration

• Add nudges and rotations

• Repeat until in a similar position to sword

My biggest concern here is the imbalance aspect. I'm planning on addressing that by just practicing both sides, but I see nothing stopping him from becoming a sith lord.

Also of note, this is just the plan I'm coming up. I may not need all of these steps, or I may need even more steps. We'll find out! So let's attack it!

Generalizing perturbations**

This was our first shot at it:

Future attempts looked more sophisticated I swear.

Here I started by applying small nudges to the pool noodle in each direction. Of note is he's starting to grab it just slightly harder each time. That's what we want!

And that's all for now! I'm going to spend some time on this because this behavior is key, and I can't really move forward without it. I envy those of you who got this for free! If you want to join in on making your own sword-dog, please do! I'd love to see your progress in the comments! If you really want to have some fun, join us over on DogKatas! We're growing and would love to see what you're up to!

See you next week, and Happy Training!

---

* I spent a lot of time looking up if this is proofing or generalization. I think that this is technically more on the generalization side of things, though the definitions of proofing seem to imply that proofing is a part of generalization, and that proofing is more about environmental stimuli.

Given that this definition doesn't really affect anything about how we're training, I'm not going to spend more energy worrying about it until it matters. Fight me.

** I wanted an excuse to use this word because it sounds cool. See? My degree wasn't totally useless!