Training a service dog requires extensive time, financial investment, and specialized expertise. Yet many dogs are ultimately released from these programs due to unsuitable behavioral characteristics. Identifying promising candidates earlier could dramatically improve efficiency and reduce training costs.
This study explored whether instrumented dog toys—a silicone ball and a silicone tug device—could capture meaningful behavioral signals during play that correlate with training outcomes. Over a two-year longitudinal period, 40 dogs entering advanced training with Canine Companions for Independence were evaluated using these sensor-equipped toys.
Researchers extracted key features from each dog’s interaction patterns and developed a logistic model tree classifier. Remarkably, using only five sensor-derived behavioral features, the model predicted whether a dog would successfully complete training with 87.5% average accuracy under randomized 10-fold cross-validation.
Further testing on a single working dog over 1.5 years demonstrated consistent predictions, underscoring the reliability of the approach. Importantly, early identification of dogs unlikely to succeed carries substantial resource implications: for a cohort of 40 dogs, the estimated savings exceed $70,000. Across all six CCI training centers, projected annual savings could surpass $5 million.
These findings highlight the potential for objective, technology-assisted assessments to support more efficient and humane service-dog training pipelines. Sensor-based behavior profiling may offer a scalable and non-intrusive tool for predicting suitability early in a dog’s development.
Source: Byrne, C., Zuerndorfer, J., et al. (2018). Predicting the Suitability of Service Animals Using Instrumented Dog Toys. Proceedings of the ACM on Interactive Mobile and Ubiquitous Technologies.
