Breakout Session 2 B

Animal training is a complex and challenging process that requires a high level of expertise and precision. Trainers develop training plans, make split-second decisions, document the dog's behavior, and adjust the training plan accordingly. To help trainers in this process, here I advocate for the use of technology, such as computer software and devices. These can assist by collecting data, making suggestions for training criteria, deciding on the accomplishment of criteria, and even performing complete training steps. The result is a data-driven, reproducible, and well-documented training that makes it easier for the trainer to focus on the specific needs of each dog.
To teach a new behavior, trainers typically develop a training plan in a word processing software, perform the training while recording performance, and document the results in a spreadsheet. This separation of tasks hinders the training loop's adaptability and slows its overall efficacy. To address this, I developed an application that integrates these tasks (http://www.plantraindoc.de). Trainers can generate training plans by repeatedly breaking down a desired behavior into smaller approximations. These can then be trained while the app records each trial's success. Thus, the training is documented continuously, and progress can be evaluated immediately using different statistics. This makes it easy to adjust the training plan in real-time with minimal interruption to the training session.
Some training steps require the introduction of a random element, like the duration of time between reinforcements or the placement of the target odor in a line-up. As humans, we're notoriously bad at generating truly random sequences. Here, the computer can assist by suggesting training criteria. These can be entirely random, as in the case of line-up, or centered around a specific mean for distance and duration. This frees the trainer from memorizing numerical sequences or using sheets of papers during the training process.
Measurement and recording of data beyond mere success rates can be achieved using sensors and cameras. The app can control a video camera and its recording is directly linked to the training steps, ensuring that the trainer can readily reference the associated video footage. Ultra-Wide-Band sensors could be used to determine and record the location of the dog in the training area. For detection dogs, an AI-based sensor capable of classifying sniffing behavior is currently in development. In addition, real-time computer vision will enable identification of behaviors like sitting, standing, or lying down. Finally, physiological data like body temperature and heart rate could be recorded. The collection of these data provides valuable information for the fine-tuning of training plans, as well as for the computational evaluation and comparison of training success between dogs.
In this setup, the computer "knows" the current criterion and can measure the corresponding data in real-time. Thus, it can also make decisions on the completion of a criterion. This results in clear criteria without any room for human interpretation, providing a black and white setup that is beneficial for both the dog and the trainer. The dog can learn more effectively as there is no ambiguity in the success criteria, and the trainer can focus on other aspects of the training process. The latter is particularly useful in medical training, where the trainer can concentrate on the procedure and the dog's reaction while the computer keeps track of cooperation signals and duration.
Finally, by adding automated reward delivery, a fully computer-driven training approach could be implemented. This can be advantageous in situations where the trainer is unavailable or prefers not to be present. For instance, the dog can be trained to remain in place even when the trainer is in a different location. In the case of detection dogs, getting the human out of the equation avoids unwanted cues from the dog handler. Furthermore, training can be standardized, leaving less room for unwanted variability.
Having training plans and data in a single app enables seamless sharing by syncing to a web server. This is useful for both, feedback on your own training by other dog handlers, as well as for supervising clients. Furthermore, sharing training plans and progress can facilitate long-term review of the training, and enable data-driven analysis of training effectiveness.
Taken together, Computer Assisted Animal Training should not replace the human trainer. Rather, it should make it easier to focus on and adapt to the most important member of the training process - the dog.