Skateboarding, or, What I Learned from the Chickens
It was December of 1999. A few of us were sitting around complaining about how hard it was to get our students to let go of food – luring. Someone came up with the idea that we might introduce skateboard training, since it would be very hard to train that behavior through any form of prompting. They would really have to shape the behavior.
I decided to try it first with my own dog (always a good idea!). I worked on it for a couple of weeks with increasing frustration for both of us. My experience with Ruby, a German shepherd, had always been that she was almost a one – trial learner. I was completely baffled by her apparent inability to learn this new behavior. Then, in January, I went to Bob and Marian Bailey’ s Chicken Camp (aka “Workshop on Operant Conditioning”). I came back, worked on it for another few weeks — and we had it!
This is the story of how that all happened, my journey into operant training, and what I learned.
Early efforts at training
Part of the frustration in the early stages of the journey came from my not really knowing the destination — in other words, the topography, or what the finished behavior would actually look like. On November 27, after a few training sessions, I wrote in my training journal: I’m not really even sure what the finished behavior will look like! I was imagining that she’d have one front foot on the board and use the other front foot to push off. Then I imagined that, once the board was moving, she’d jump onto it with all four feet.
I had started the process by having Ruby play the Box Game using the skateboard as our box. She had played that game many times before, with other objects, and she had no trouble at all doing various things with the skateboard — looking at it, nosing it, putting a paw on it. My only attempt at an analysis of the behavior consisted of waiting to see how she approached the problem — as if she knew that she was supposed to learn to ride the thing!
When I saw that we were going nowhere, I interrupted our training and consulted with some skateboard – training experts (there are just a few in North America!). I also watched video segments of dogs of different sizes riding skateboards. Finally, I was able to describe the goal behavior: Ruby will put both front feet on one end of the board and, by pushing off with her hind feet, she will move the board forward.
It was no problem at all to teach her to put both front feet on the board, though we did muddle through a few sessions in which I trained her to put one foot on the board and hold the other one up in the air, on the way to the board! By this time, though, I had reinforced enough of her approaches to the middle, instead of the end, of the board that she was consistently approaching the board from the left and putting both front feet on the middle. When I tried to extinguish that by not reinforcing the behavior, she reverted to her default Down with both front feet on the middle of the board. Oh my! I always knew she was a quick study, and now I knew what an excellent trainer I was! That old expression “You get what you train” rang in my ears.
I decided I had to split the behavior she was giving me and isolate the Down part of it. I noticed that her head dropped slightly whenever she started to lie down, so I began to click her for front feet on the board (not worrying at this point that they were on the middle) just before she dropped her head.
Whenever I was too late and the Down was already in progress, I simply waited until she began to rise out of the Down and clicked that movement. The result was a beautifully trained, regal statue: front feet on the middle of the board and back arched! Picture the carved wooden figure on the prow of an old sailing ship. That was Ruby!
After Chicken Camp
That was where we were when I went to Chicken Camp in January 2000. I spent the better part of a delicious dinner with Bob and Marian Bailey discussing the problem. Somehow, Bob managed to keep a straight face as I described what we had done so far, and then he asked if I wanted to know how he would have done it. Yes! He described his universal strategy for training a complex behavior. To paraphrase:
When you’re training a complex behavior, first decide what is the essential part of that behavior. Then, modify the environment so as to isolate that part. The idea is to make the learning as easy as possible for the animal. Shape that essential part (“cheating” as necessary, for example by luring). Then, gradually change the environment back to what it will be when the behavior is complete. So, in training the skateboard riding, start by building a facsimile “skateboard” with wheels and just enough platform for the dog to put her two front paws on. Then shape that response. Once that response is trained, gradually modify the environment back to its natural state. The small platform with wheels will gradually become a full-blown skateboard. Remember: A good trainer works hard to make the learning process as easy as possible for the trainee.
I returned home after Chicken Camp with renewed inspiration. Over the ensuing few weeks of training, I struggled with arranging the environment — something I had never really thought about much before this training challenge that I had set for myself. My task was complicated by the fact that Ruby now had a very solid Statue behavior that was off by 90 degrees. I worked very hard, from that point on, to make it next to impossible for her to approach the skateboard from the left side and land on the middle of it. If not for having to undo my earlier training mistake, I could have accomplished my goal in just a few training sessions.
I became very inventive about making only the very end of the skateboard available to Ruby, and I stabilized the board so that it wouldn’t move when she put her paws on it. I covered all but the end of the board with a low table. I positioned the board in a corner of the room and carefully arranged small trash cans on both sides of the board. Then I faded and finally removed one, and then both, of the trash cans. Then I gradually moved the board away from the corner, one foot at a time, so that both sides were accessible. For a person who is normally very intuitive and global in her thinking, I became painfully concrete.
As my powers of focused observation grew, I became aware of how my own position in the room was influencing Ruby’s angle of approach to the skateboard. She was always trying to face me. Not surprising, since that had always been part of our clicker dialogue. When I’m shaping a behavior, she offers a response and then looks at me, as if to say, “ Is that what you want?” So, once I was satisfied that I had arranged the apparatus itself for maximum success, I then worked on my own position. At first, I positioned myself so that she could see me only if she approached the board correctly, from the end. Then, I gradually faded my position as her prompt until I was able to stand about 15 feet directly behind her.
Later on in the process, when I had the skateboard out in the middle of the room, I used my body position and movement to prompt the desired behavior by straddling the opposite end of the board to block anything but a rear approach, and by running her up to the end of the board.
Back when the board was still positioned in a corner of the room, I began to notice that I had really been working on two criteria at the same time: Ruby’s movement towards the end of the board, and her angle of approach. I had bee n tossing treats at varying distances behind Ruby to reset the behavior, but that simply made it easier for her to go into motion and harder for her to concentrate on angle. Since I wanted to work on the angle criterion at this point, I began to toss the treats right behind her. All she had to do was give the response, spin around in place to get the treat, and then be in position to give the next response. I also used a very high rate of reinforcement whenever her angle of approach was perfect. I’d simply keep her feet on the board by quickly handing her treat after treat in position: another lesson learned from the chickens about strategic delivery of treats!
I spent some time isolating the angle criterion by going through the same process of stimulus reversal that we had used with our chickens. The left side of the board was Ruby’s previous hot target that I wanted to extinguish. So I began to click anything other than a left-side approach. Then, I began to shape the behavior I wanted: the approach from the rear. Sometimes, she would follow her left-side (unreinforced) approach with a right – side (also unreinforced) approach. It took only a few short training sessions before she began to correct her own approach. The first time this happened, she approached from the left and then, without even waiting for the non-click, she swung herself around to the middle. We had it!
Ironically, I realized much later in the process that this was really a false criterion. It would have been better to leave that decision to Ruby and allow her to make it in the moment, depending on the immediate circumstances. I had not taken into account the efficiency of natural, untrained behavior and so had wasted time working on a criterion that was really irrelevant. The work we did on this criterion was useful only as a way to compensate for my earlier mistake in training Ruby to approach the board from the middle. Once that problem was solved, there was no longer any need to keep pushing on the angle-of-approach criterion.
Gaining and using momentum
After about six days of training following my return from Chicken Camp, I suddenly became impatient with the careful and painstaking process I had established. For the next dozen or so trials, I experimented with various forms of prompting, combining food – luring with my own movement. The skateboard was still in the corner of the room. I started by walking her up to the end of the board following a treat in my hand. After just a few lured repetitions, I began using the treat as a reinforcer, tossing it after the click.
Then I backed us up a few steps and ran us up to the rear of the board, in effect modeling the response I wanted. At this point, she was getting 100% accuracy even when I started hanging back a step (or two or three) and letting her complete the run on her own. I gradually increased the distance of the run to about 6 feet, tossing the treats in a straight line behind her.
We got to the point where I only had to shift my body weight forward to start her running. A couple of times, she even performed this running approach without any help from me, when the treat fell close to the board. Instead of starting her back again for her run, I just let her work it out for herself, and she did! She got her treat and then promptly ran to the rear of the board and put her front feet on it.
I knew that I was lumping dreadfully, working on several different criteria at once: run to the board, approach it from the rear, and put your front feet on it. But it was working! Part of the explanation, of course, lies in the fact that she had probably mastered two of the three criteria by now (the rear approach and the front feet on the board). But I believe that the rapid movement was also an important part of the training process. I wrote in my training journal: “It might just be that some responses are better trained with movement, rather than splitting them down to their tiniest approximations.”
Or it might just be quite simple and pragmatic: If you’re moving fast, then you’re most likely to keep moving fast — following the principle that animals try to get the most reward for the least effort. That’s like what we do when we teach fast recalls to beginning students. When a dog is moving fast, it’s less likely that they will be distracted along the way.
Also, the movement (especially fast movement) seemed to be energizing and reinforcing to her. I’m thinking back to Marian Bailey’s explanation of t he autonomic nervous system and wondering if that has something to do with it. Is expending energy (activating the sympathetic nervous system) a way of reducing stress (mediated by the parasympathetic nervous system)? Was the fast movement actually shifting Ruby out of the sympathetic into the parasympathetic and thereby optimizing her ability to learn?
Following this session, we had a four – day gap in training when I was out of town. Upon my return, I discovered that Ruby had lost none of her learning. Aft er a few review sessions, my intuition told me to make a huge leap in criteria, and I moved the board out into the middle of the room. I continued to use my body position as a prompt to keep Ruby approaching from the rear, and I used my feet to control the movement of the skateboard itself. I allowed the board to roll just a few inches toward me, until I was certain that the movement of the board was not disturbing or distracting to Ruby.
Then I took a few steps back from the board, allowing Ruby to slide it toward me. She began to mount the board from either end, depending on which was closest to her when she picked up the treat I had tossed.
In order to lengthen the duration of the ride, I began to delay the click, and she began surfing! She’d put her front feet and chest on the board, as she did when going into a Down position, and sort of glide a few feet, without actually pushing or walking with her hind feet(the lazy German shepherd approach to skateboard riding!). She also began moving her hind feet to push off and start the skateboard in motion.
A few days later, we had an opportunity to show off at a meeting of our local clicker group. The meeting was held in a different building and attended by about 20 people and several dogs, most of them strangers to Ruby, seated in a big circle around the perimeter of the room. I put the skateboard on the floor, and Ruby immediately ran to it, put her front feet on the end, and slid it a few steps forward.
I’m still working on choosing an appropriate cue. Our current working cue is, “Ride ‘em, cowgirl!”
Analyze and plan. Before you start training, have a clear picture of the topography of the finished behavior. You can’t get there if you don t know where you’re going!
Specify the ladder of approximations that will get you there. Be sure that each step is small enough that the dog will be successful.
Become an excellent observer. You can’t shape a behavior if you can’t see the responses your dog offers along the way.
Train only one criterion at a time. If you work on more than one, you’ll only confuse the dog, learning will be slow, and you may even lose your dog along the way.
Work in short sessions, of a length that’s appropriate to the dog and the task, and evaluate progress before starting the next session.
Keep written records to help you move most efficiently toward the goal behavior. A training log will also help you retain the learning that you got out of the experience. And it will keep you going through the rough times. The trainer, too, needs positive reinforcement!
Testing is part of training. And if you’re testing for readiness to move on to the next criterion, don’t think in terms of a pass or fail test, but simply an information-gathering event that will help you decide on your next step.
Get the behavior. Once you decide on a criterion to work on, your timing must be impeccable so as to send clear information to the dog. It also helps to reinforce movement rather than static position.
Keep the rate of reinforcement high enough to keep your dog in the game. When necessary, use a high rate of reinforcement to emphasize a correct response.
Think strategically about how and where to deliver the reinforcer, whether to hand it to the dog or toss it (and in which direction and how far away).
Prompt the response, if necessary, by luring with food, by making use of the physical environment, or by using your own body position and movement. Use whatever it takes to ensure the dog’s success.
Decide when and how to split or lump. Don’t get stuck on a particular split just to follow your plan.
Keep the big picture — the goal behavior — clearly in mind at all times, and work steadily towards it. It’s tempting and reinforcing to the trainer to focus in on the tiny details, and on the beauty of the shaping process itself, and it’s easy to get stuck there!
Stay open to what your intuition tells you and be ready to make the occasional leap of faith. That’s the art part of training.
Remember the basic principles of operant conditioning. You get what you train. Behaviors are built through reinforcement. Understand and appreciate the power of non-reinforcement to a dog with a strong history of reinforcement.
If you don’t like what you’ve got, put it through a process of extinction first, by reinforcing anything that’s the undesired response, and then by shaping the desired one.
Develop stimulus generalization in gradual increments. Don’t jump from a skateboard surrounded by barriers to asking your dog to compete in a skateboard-riding event!
One of the really interesting and challenging things about this skateboard problem is that each dog is likely to do it differently. There’s no right way to do it. So, for the trainer, there has to be a lot of moving back and forth between criteria, “listening” to what the dog is telling you, making appropriate adjustments, not getting too hung up on one thing when it seems not to be advancing the process, etc. A great learning experience!