Our lunch session today consisted of some precision work for heeling and retrieve practice.
Over the last two months Perrin has progressed from right side pivots on a platform, to off platform pivots and some short stretches of heeling on the right with lots of turns. Through this, I have been heavily rewarding engagement, enthusiasm, and closeness to me. Because of my priority choices, I have lost some of the awesome muscle memory for straightness that Perrin had developed through the pivot work, so I have decided to go back and brush up on that skill for a bit. His muscle memory for right turns is very good, and I also want to go back and develop the same for left turns, as well as begin on some left side heeling foundations.
We also worked on retrieves for a bit. I am working on putting the retrieval of my car keys on cue (‘Keys’). We did some basic retrieves, and I am using the standard protocol for putting a shaped behaviour on cue I learned in one of Donna Hill’s classes, and at the moment we are on step one: set up the environment, say the cue as the dog does the behaviour for ~50 reps. I also started adding some distraction, but I really should have waited on that until I have the cue stronger. I shouldn’t be trying to mix those two right now. I could feel the retrieve behaviour eroding every time he brought me back the wrong object and he didn’t get reinforcement; his retrieve isn’t a heavily reinforced enough behaviour yet. However because I was trying to put ‘Keys’ on cue, I also didn’t want to confuse the meaning of the cue to mean ‘retrieve anything’ rather than ‘retrieve the keys’. I didn’t put enough thought into that up front.
It did give me a good idea for another differential reinforcement exercise to set up though and I think I will try it for tonight’s session. I like the idea for this stage of learning because then he still gets something for completing a great retrieve but gets something better for retrieving the object I want.
In my head it looks something like this:
- set up a number of retrievable objects in the work area (dog toys, brushes, keys, mittens, socks, containers etc)
- Have a low value and a high value reinforcer (in this case it will be my standard homemade training treats and some cheese)
- Pick a ‘target’ object (in my case it will be the keys)
- For every non-target retrieved to hand, give a low cookie. For every target object give cheese
- Retrieves that are dropped, or otherwise don’t make it to my hand are not reinforced.
- I will not use a cue unless/until he is consistently getting the keys, so I don’t confuse the meaning of something I am still working on putting on cue.
I tried to set that plan up tonight and it didn’t go the way I expected at all! Perrin never brought me the wrong item!
Because Perrin had brought me different items during our lunch session, I just assumed he would do the same later. I forgot to take into account the reinforcement history that Perrin has for retrieving my keys, especially when I had never asked him to retrieve any of the other items in the pile before (so zero reinforcement history). It was mediocre retrieve training, but a bad experiment. To make it better retrieve training, I needed to wait until all objects are stationary before sending Perrin out to retrieve. For much of the video, he is just going after the object that had most recently moved. When I did a better job of setting him up, he searched a bit longer (but still brought back the right one!).
If I wanted to test that theory more cleanly, I should choose a pile of objects that have a similar reinforcement history (most easily achieved by using all novel items). Perhaps tomorrow.
ETA: I just thought of another potential issue with that experiment design: Perrin goes and picks an item out of a pile of novel objects and retrievs it. Then he gets a low value cookie, and then continues to just bring me that object. If that happened, he would be unaware that a higher value reward was available, and would be quite happy to keep working for the low value cookie. If I were to make whatever object Perrin brought to me first the ‘target’ object, I may end up in the same place, but just giving cheese rather than cookies. I will have to set this up and see what happens, and go from there.