Acceleration-based Clustering Reveals Frequent Gait Switching in Sprint Sled Dogs

Area of Science:

• Animal locomotion biomechanics
• Canine physiology
• Data science in sports analysis

Background:

• Continuous video analysis of sprint sled dog gallop is challenging in field studies.
• Understanding stride-to-stride variation is crucial for optimizing performance and welfare.

Purpose of the Study:

• To develop and validate an acceleration-based pipeline for identifying stride states in sprint sled dogs without manual gait labels.
• To analyze stride variation and transitions during load-pulling gallop in a team setting.

Main Methods:

• Utilized harness-mounted tri-axial accelerometers to collect stride data.
• Applied multivariate dynamic time warping, manifold embedding, and density-based clustering to over 20,000 strides.
• Correlated acceleration-derived states with manually labeled gallop patterns in a subset of data.

Main Results:

• Identified recurrent, dog-specific stride states from acceleration data.
• Observed frequent transitions between stride states with significant interindividual variation.
• Found limited evidence of strong team-level coordination in stride state transitions.
• Demonstrated weak predictive power of local tugline-force timing and magnitude for transition events.

Conclusions:

• Sprint sled dog gallop involves a dynamic set of stride states rather than a single pattern.
• Local tugline-force fluctuations alone do not adequately explain the observed stride state switching.
• The developed acceleration-based pipeline offers a viable alternative for gait analysis in challenging field conditions.