At the Art and Science of Animal Training conference this year Alexandra Kurland spoke about loopy training, which is a concept she has been developing over the past year. Loopy training was the focus of a clinic I did with Alexandra Kurland last fall. The more I hear about it, the more it makes sense and the more I’m able to see how to apply it to make me a better trainer. This post is a combination of my notes from the conference and from the fall 2009 Alvin clinic.
The great thing about loopy training is that you’re probably already using it! However, the loopy training model gives us a clear framework to look at training. Loopy training helps us see what makes great training great and helps us figure out what’s going on when problems occur. If we can formalize and name concepts (such as the loopy training concept), it makes the training process more deliberate, simpler to teach and easier to discuss and assess.
Basics of Good Training (just so we’re all on the same page!)
Animal training (as Bob Bailey often says) is a mechanical skill. We shouldn’t let poor mechanics get in the way of good training. Alexandra Kurland shared three things that contribute to great training:
- Clear, well, defined criteria
- Great timing
- High rates of reinforcement
In addition, good training requires that we are splitters, not lumpers. Being a splitter means that when teaching, we break down behaviors into small enough pieces so that the animal can be successful.
Alexandra Kurland talked some about poisoned cues, which can occur when we combine positive reinforcement with negative reinforcement. (I’m going to save poisoned cues for another post in an effort to keep this one a reasonable length!) Basically, though, when we use aversives or when we use negative reinforcement poorly, the emotional effects linger long after the use of corrections has stopped. Clicker training can bypass the poisoned cue effect by retraining old behaviors in new ways.
We want happy, exuberant, eager to please animals and this is what we usually see when training with positive reinforcement. It’s important to retrain familiar behaviors even if the horse already “knows that” because these simple behaviors become anchor behaviors as we build more complex behaviors and loops. If we build a strong foundation, cues for simple behaviors can be used to reinforce and build more complex behaviors.
By now you’re wondering…what the heck is loopy training?
We generally think of behavior using the following model:
Cue –> Behavior –> Click –> Deliver Reinforcer
However, this is too simple of a model. This is how training really looks:
Cue –> Behavior –> Click –> Deliver Reinforcer –> Cue –> Behavior –> Click –> Deliver Reinforcer –> Cue –> Behavior –> Click –> Deliver Reinforcer –> Cue –> Behavior –> Click –> Deliver Reinforcer
Behaviors and training happens as a LOOP, rather than as a single instance in time. Also, although we might be thinking only of the cue we’re giving and the behavior we want to reinforce, there’s actually a lot more behavior going on! The goal in training is to create clean loops.
A clean loop will have the following features:
- No unwanted behaviors in the loop
- Animal performs all elements smoothly
- Behaviors in the loop are performed with no hesitation
The framework of loopy training emphasizes that all elements in a loop need to be clean, both before and after the click. If you have a clean loop, you can (and should!) move on. If you have a behavior loop that stays messy and does not seem to be getting cleaner, reassess your training. Are you asking for too much? Have you skipped a step? How can you better motivate your horse or improve communication?
The Click as a Cue
The click serves several different functions and ends up getting called all sorts of different names, a marker signal, a yes answer signal, a conditioned reinforcer, a bridging signal, a highlighter.
Most importantly, the clicker, if used properly, functions as a cue, or a green light for earning reinforcement. This idea came out of a research project at the University of North Texas. In that study, the trainer had a nice clean loop with a dog touching a target across the room. She delivered treats by tossing them into a bowl at her feet. However, in order to increase precision, she changed from tossing the treats to dropping them down a piece of PVC pipe.
Her clean loop fell apart completely. After the click and food delivery, the dog stopped, looked at her, paused and then repeated the behavior. The dog made several more attempts and looked quite frustrated before giving up all together! In this situation, the click was serving as a cue to look at the handler and watch for the toss and landing of the food. The dog stopped and looked, but because of the new PVC pipe, no toss happened. The poor confused dog had no idea what was going on.
Feeding skills and food delivery are essential for good training! The animal must know when and how food is delivered and must be able to collect the food promptly, efficiently and safely. We can have different ways of delivering food (tossing on the ground, tossing to the animal, giving to the animal, etc.) but we must train each and make sure the food delivery part of the loop is always clean. When horses mug for treats, it often indicates that either they have not been taught proper manners and self control around food or that our food delivery is inefficient or clumsy.
Inefficient, awkward or clumsy food delivery will break down the smooth flow of the loop. I watched a DVD last semester where the trainer was delivering food by tossing it on the floor. The dog did not know how to watch the throw and find the food. So, there was lots of extra sniffing and searching behavior and the pace of the training was often interrupted. The trainer eventually got the behavior, but the process was neither efficient, nor elegant.
What are your foundation lessons?
Good training starts with small chunks. If we can get small, clean chunks of behavior then it is easy to start combining to get more complex patterns.
Once we have clean loops for at least two to three behaviors its easy to start mixing and matching. Having a clean loop does not mean the behavior is perfect, it just means the cycle is fluid without any unwanted behavior. Once we build an inventory of simple loops we can start on more complicated loops. Importantly, we can also use the cues and behaviors in our simple loops to reinforce other behaviors we are working on.
Alexandra has six basic foundation lessons she teaches her horses. This is a way to introduce the horse to training and start building some simple loops. The foundation lessons she selects are important because they teach behaviors related to self control and safety, as well as being behaviors that can be expanded later to teach all sorts of other behaviors.
Having organized foundation lessons means that you have a plan for training. You understand what skills are essential to teach the animal early on and you recognize how these skills can be combined or developed to teach more advanced behaviors. Foundations skills are like loops. Very likely, you already use them, but you’ve never formalized them.
So I have a clean loop, now what?
Here’s a nice simple loop that you might see in horse training (or dog training):
cue –> horse backs up several steps –> click –> deliver reinforcer –> cue –> horse backs up several steps –> click –> deliver reinforcer –> cue –> and so on!
How can we use this to our advantage in training?
In our loopy model, cues serve a double function. A cue can:
- 1. Tell the animal what next behavior will earn reinforcement.
- 2. Reinforce whatever behavior occurred previously.
So, later on we might have:
cue 1 –> horse backs up several steps –> cue 2–> horse lowers her head–> click –> deliver reinforcer –> cue 1 –> horse backs up several steps –> cue 2 –> horse lovers her head –> click –> deliver reinforcer –> cue 1 –> and so on!
If both behaviors started out as clean loops, we can easily combine them into a larger loop. More importantly, we can use the cue for head lowering to reinforce really good backing up. So, we ask for the horse to back up. When we get several really good steps of backing up, we ask for head lowering. Since the head lowering cue signals a chance to earn reinforcement, the cue itself will reinforce good backing up.
As the animal learns more behaviors, we can combine them in all sorts of ways and patterns until we end up with quite complex loops. We might have head lowering then backing then head lowering then a click and treat, then a hindquarter yield and then backing and then click treat then 5 target touches and repeat. Clean training is built on loops, even if it might be difficult to see them at the more advanced and complex stages. When coming up with a training plan at more complex stages of training, think about where you need the primary reinforcement and what the horse is telling you he needs to work on.
Loopy Training, Poisoned Cues and Negative Reinforcement
Loopy training is the antidote for poisoned cues because poisoned cues are a product of lumping. When we teach behavior too quickly or in too big of chunks, we often end up with frustrated, annoyed animals who are not enthusiastic about training. These poisoned cues don’t work well in loops! We can avoid frustration and unwanted behavior by building tight, clean loops that gradually spiral outward.
Most of the time when we use negative reinforcement we’re guilty of lumping. It’s not the negative reinforcement that poisons the cues, but the lumping that creates stress, frustration and confusion for our horses. We can create equally frustrated animals if we’re using positive reinforcement, but lumping behaviors together rather than splitting and shaping gradually. Now, this doesn’t mean train in the tiniest steps possible. It only means we must progress at a level where the animal can continue to be a successful learner.
Later on, we can also use loops to give our horses a choice or to let them cue us. What is very important to animals is how much control they have over the environment. We can make our animals more confident and comfortable by giving them more control. For instance, you could teach your horse to touch a target before you put on the saddle. The horse then learns he can use the target touch to tell you when he’s ready to for you to saddle him. If the horse later on chooses not to touch the target, then it’s up to you to figure out why he doesn’t want to be saddled that day.
Loopy training helps us be more deliberate as trainers. Behavior happens as a cycle, or loop, rather than as a discrete unit. If we can see these loops and learn how to use cues more effectively, our training and animals will benefit. At the core, loops work because they teach us to split behavior in small chunks and get good quality behavior before trying to move on.
If you enjoyed this post, you might enjoy the rest of my notes from the 2010 Art and Science of Animal Training Conference or my notes from the 2009 conference. Better yet, bookmark the ORCA website and come to the 2011 conference next spring! Sign up for e-mail updates to make sure you don’t miss any of the great posts from stalecheerios.com.