Teaching quake safety—with dogs!
Opportunities come by many routes. Recently, I received a phone call from an organization preparing a group of dogs to help teach earthquake safety to schoolchildren. Just as dogs can demonstrate the stop, drop, and roll of fire safety, these dogs would be trained to demonstrate how to drop to the ground, crawl beneath sturdy furniture, and grab on for stability—basic earthquake-safety procedures.
Working with these dogs sounded like a fun challenge. Not only could I refresh my own earthquake-safety knowledge (living along the New Madrid fault, we tend to ignore earthquake preparations because quakes are either fairly minor or have the potential to destroy us all anyway), but I could also create a great new behavior.
I knew I needed to develop a clear and simple training plan. I would have only a single session to teach the new behavior to the group, and the group would have only two weeks to practice. Before we met in the training session, I would not have met any of the dogs or handlers, and would have no sense of their foundation skills. These circumstances underlined the need for a straightforward, well-thought-out plan. As I made that plan, I decided that sending a video home with the learners after our only class would provide additional value for our short time together.
Shakespeare, my 11-year-old Doberman, and I began working out a sample plan—and the video. Shakespeare is quick to remind me that plans are organic things that don’t always go… according to plan! Nevertheless, there is no substitute for a thoughtful training plan, this one supplemented by a video takeaway.
Plan started, memories sparked
Shakespeare loves shaping, but I don’t use target sticks often. My dogs are tall, and I generally use few props. I had used a target stick only rarely with Shakespeare, my first clicker dog. But, I knew he would have little trouble picking up the concept of a target stick. I brought out a target stick and began clicking for foot movement.
Hmm. I wasn’t getting much foot movement. I got a lot of nose targeting, but not much paw. This perplexed me. I should have been starting with a blank slate, and we’ve done both nose and paw targeting before. Why was Shakespeare so insistent on using his nose now?
I stopped—not nearly as soon as I should have—and thought back for a moment. Oh, yes! I had used a target stick with Shakespeare at an event with Terry Ryan to teach a chin-to-ground behavior. I’d shaped it with Laev, my other dog, but I had used the target stick with Shakespeare. That was quite a while back, wasn’t it?
But clicker training is remarkably persistent in the brain. It didn’t matter that I had trained Shakespeare with the target stick for a silly trick 2300 miles away and eight years earlier! Shakespeare remembered the target stick and couldn’t understand why I did not. I chose a new paw target, a piece of junk mail, free from all previous associations, started again, and saw immediate success. As B.F. Skinner said, “The rat is always right.” My dog did exactly what he had been trained to do. I was reminded that if something seems unreasonable during training, it probably means that I misunderstood or missed a step.
Refining the base behavior, the paw target
My goal was a solid target so that I could place the paw exactly on the furniture, but Shakespeare was keeping his face pretty near the foot target (a holdover from early agility training—again, the rat is always right!). So that Shakespeare wouldn’t bring the head movement to the furniture behavior, I refined the paw target, moving it so that Shakespeare had to concentrate on touching only with his foot.
Shakespeare responded enthusiastically. I was saved from refining the behavior further by his unintentional self-punishment when he tipped backward. He was more cautious during the next repetition, I was able to click the smaller movement, and Shakespeare realized that the smaller movement was all that was necessary.
In our training video, there is one time when I removed the target because I saw Shakespeare starting to nose it instead of pawing it. I withdrew the target, preventing that mistake (and also imparting a negative punishment), and then offered another chance. I did not use a no-reward-marker; it was obvious to Shakespeare that the nose experiment hadn’t worked!
To set up Shakespeare for the correct behavior, I cued a sit before offering the target. I was able to do this only because I was sure I would get a paw target upon presentation of the target (the cue). If Shakespeare had been uncertain of the paw target behavior or cue, I would not have cued a sit. Following a known cue with an uncertain cue creates a very weak behavior chain. In this example, that error would have weakened the sit behavior in this context (and possibly elsewhere, depending on the dog’s experience). Keep your cues and chains clean.
The perfect position is hard to find
Before moving on, I had to be sure Shakespeare was comfortable targeting from a down position, as the earthquake-demo dogs would be crawling toward this final behavior. We had already practiced in several angles and positions, so Shakespeare had no problem with the new position.
In the video I use the target as the cue, presenting the stick when I want the behavior. Because I’m not yet working with the final behavior (grasping the furniture rather than pawing the junk mail), I don’t attach any other kind of cue yet. There’s no reason to take the time to do that at this stage (more on that later!).
Next, I wanted Shakespeare to straddle the chair leg. Since he was comfortable targeting with either paw, I assumed he would use the nearest paw and that he would recognize that he was ending up on either side of the chair leg. As it turned out, that was an unrealistic goal for a project that needed speedy learning—and it was not an assumption I could make for the unknown demo dogs. Shakespeare could learn to target that way eventually, but ease and speed were the requirements for my shaping plan. We needed a change.
(Why did I click when he stood to target, the second rep in the video clip? It was a judgment call, and I chose to click. Targeting-while-down was my goal, but this was only his second attempt, with new aspects to the behavior [working around the chair]. Shakespeare was correct in the paw target, and I knew that he wouldn’t take long to look for the most efficient route to his click—moving only the leg instead of standing.
If I had opted not to click, I would have had to remove the target as he stood so that he could not complete the behavior without a click. To allow Shakespeare to complete the behavior without a click would have been confusing, because the targeting behavior was actually correct! If standing was the mistake, I could not let him perform a correct behavior afterward without reinforcement.)
For simplicity and speed, I changed the plan. I would use a paw target to wrap Shakespeare’s leg about the furniture gradually rather than have him straddle the furniture.
I needed a more useful target, one that could stick to the furniture rather than one that required me to present it, so I transferred the target from the junk mail to a yellow sticky note.
In the video, you can see that Shakespeare reverts to sloppy touches and some nose touches (and you can catch several errors on my part!), but I don’t particularly mind at this point. Why? I’m not sure exactly what motions Shakespeare will need to get his paw around the chair leg, so I don’t want to be too limiting with my target work. He’s reaching toward it—that’s good enough for what I need. Since the behavior is not yet associated with a verbal cue, I can clean up the behavior before attaching a proper cue.
If I’d wanted cleaner behavior, I’d need cleaner clicks. That would mean I’d have to change my viewing angle, which was partially obstructed by Shakespeare’s own body. As it turns out, that was no big deal here, but it could be a major deal in another training session! Consider your own abilities when setting up your training.
As I bent to place the target in its new location, Shakespeare recognized the target—his cue—and acted. I didn’t have my clicker ready (it was in my hand, but my thumb wasn’t on it) and so I marked the behavior verbally and reinforced it.
If I had not reinforced, Shakespeare’s first attempt at the behavior with this new criteria (on the chair) would have been unsuccessful, and he may have concluded that his behavior was incorrect in this scenario. We could have fixed that, but the cost would have been lost time and energy.
Note that in the second rep in this video, I click for just a lift of the paw toward the chair. Even though Shakespeare had been successful the first time, it’s still a new scenario and I’m working with looser criteria. The click provides enough information to reassure him about the targeting behavior.
I get a lot of variability in Shakespeare’s targeting behavior on the chair because I hadn’t been terribly strict about my targeting precision earlier. It is not a huge surprise when Shakespeare rests his wrist on the chair crossbar rather than targeting precisely. That is exactly the behavior I want! Shakespeare targets the crossbar, not the vertical chair leg, and it was accidental, not deliberate, and it was only vaguely influenced by the target (it didn’t occur as he reached around for a target at the back of the chair leg), but it’s the behavior I want!
Too often I see handlers refuse to click for a leap forward in a behavior chain. “He didn’t do it on purpose,” they might say, or, “I don’t know if he’s really thinking about it or if he just happened to do it.” We don’t care—it’s the behavior we want. If we mark and reinforce that behavior, it will become deliberate, even if it didn’t start out that way.
If the dog comes up with a better plan than the trainer, go with the dog!
When it was apparent that Shakespeare was comfortable resting his leg on the chair, I removed the sticky note target to avoid competing cues and started shaping the wrist grab. I took any attempt at first, but very quickly moved to clicking only for solid fetlock contact on the chair.
Shakespeare liked this plan. He liked it so much that he kept working while I was refilling my treats, though normally he will wait for me to be ready again. Again, he made a breakthrough—this time with duration—when I wasn’t ready. Again, I marked and reinforced!
I loved that Shakespeare began wrapping hard enough to pull the chair across the floor, but that behavior probably would not work with the earthquake demo, so I had to eliminate it. This was accomplished easily by clicking a bit sooner, interrupting the pull. A no-reward-marker was not necessary.
Begin stimulus control
With the good wrist wrap, we were ready to get some control of the behavior. It was time to add a cue! However, we were not yet where we wanted to be for the finished behavior, so the cue could not be the final cue. It would be an intermediate cue to be disposed of later.
I won’t cover stimulus control in this article, but in training the cue was attached in the usual fashion, first predicting the behavior, and then clicking only cued behaviors, and then offering the cue only to reinforce calm waiting.
Why did I change the cue after the first two introductory reps? I found “grab” harder to say than I’d thought, so I changed the cue to “grab it,” which flowed better. Play with your prospective cues before introducing them to your dog!
Why did I click the un-cued behavior once? Because I screwed up! Handlers and trainers make mistakes, too. Just like with the dogs, move on and quickly replace mistakes with other behaviors.
Generalizing the behavior
With some control over the behavior, I could start asking for it elsewhere. I introduced another piece of furniture, with considerably different legs, and cued the behavior. As this was a challenge, with such a different prop, I took any attempt at the behavior. Soon, though, I raised criteria back to a fetlock wrap and short hold.
Then came the toughest step: I introduced a chair without any horizontal pieces below the seat. Since Shakespeare had preferred the horizontal leg parts before (after all, that was easier), I knew this would be a challenge for him.
Again, I took any attempt at the start. Well, almost any attempt. I made the snap decision not to click his first offer—grabbing the seat. When the first rep on a new prop wasn’t clicked, Shakespeare got a bit worried and hesitated to try again. I re-cued, and he performed the behavior correctly. Be aware of this possibility; know your dog’s experience and persistence.
When I saw Shakespeare struggling to adapt the behavior, I stopped cuing and just let him experiment. I didn’t want him to worry about stimulus control or anything else while he was concentrating on balancing and on where the vertical leg would hold him.
Rest for the weary
I should have given Shakespeare a break during this last challenge, as he had been working steadily. The entire plan was a single massive session, broken up only by my treat refills and rearrangement of furniture. While Shakespeare loves to work and often doesn’t want breaks, this exercise was tough. When Shakespeare broke off and wandered away, which was very rare, and a sure sign of brain overload, I did not pressure him to return. If he’d remained unengaged, I would have quit the training, especially for a behavior I wanted to keep in his repertoire. But Shakespeare came back to work, and we kept going.
I want to emphasize that working so long and so hard in one session isn’t ideal, but I know my dog well enough to risk it. Shakespeare and I have a decade of clicking together to fall back on, and he’s unusual in his enjoyment of ultra-long sessions. For almost all other situations, I’d recommend much shorter sessions and more frequent breaks. Don’t ever get to the point where your dog needs to disengage.
During this last scenario I saw a much lower rate of reinforcement. This was at the end of a long session and during the hardest challenge. For better training I could have reduced criteria or, even better, I could scheduled a break. Again, I knew my dog would stay in the game.
With this training I was just experimenting with the concept and didn’t care if Shakespeare kept it in his repertoire perfectly. If I were training something that really mattered, I would take more care to ensure that Shakespeare didn’t get discouraged or frustrated. But even with his lower rate of reinforcement, he stays on task. That’s our strong history of reinforcement paying off!
It’s important to note that Shakespeare thinks clicker training is awesome! Even during the hardest parts, while he is panting with mental exertion, his little docked tail is wagging. He’s challenged, but he’s having fun! It’s still a game for him.
Ready to push on
When Shakespeare was more comfortable with the new chair leg, I reintroduced the cue. In the video, watch his toes as he grasps the leg now—he’s really wrapping and flexing!
We quit there, but the next step is obvious: going back to a less-challenging chair and cuing from a down position. From there we could cue with a more-challenging piece of furniture while in the down, then attach the performance cue, and, finally, back-chain the complete demo of drop, crawl, and grab.
In the end, I didn’t meet with the earthquake-demo dogs, but I am glad I had the opportunity to record the organic development of a real-life training plan. Plans are always subject to change, depending on the dog’s inclinations and progress. Even with my on-the-fly adaptations, I made many mistakes in this plan—I hope you can learn from them as you develop your own plans!