03/06/2026
We've made an exciting breakthrough in developing an algorithm capable of assessing the AFL drop punt skill.
We've successfully integrated hand landmark detection and object detection into our architecture, enabling our system to simultaneously track 21 hand landmarks per hand and read ball orientation, descent path, and contact geometry.
This means Huddl can now assess not just where the body is, but how the hands interact with the ball at every phase of the skill, bringing a level of detail that hasn't been available to PE teachers before.
Our algorithm analyses 11 biomechanical metrics across four assessed phases, with every measurement expressed as a computable value derived from landmark coordinates. Items like the grip axis angle, bilateral symmetry ratio, channel maintenance, orientation stability during descent, ball orientation at release, contact zone index and knee drive during follow through. All metrics are research-backed and grounded in Fitts and Posner's 3 stage model of motor skill acquisition, ensuring every benchmark reflects what is biomechanically achievable and developmentally appropriate for the student.
An important note is that our benchmarks are not fixed. They adapt across five developmental year bands from Pre-Primary through to Year 10. Through our research, we recognised the nuanced patterns of movement relevant to each age band. A PP student will not be capable of demonstrating a controlled 1 handed guide & release in the same capacity as a grade 10 student. Therefore, we reviewed every metric across each year band to ensure they were age appropriate, educationally relevant & research informed.
We've also built the capacity to detect and account for every meaningful variation in how a student might attempt the skill across key developmental stages. From pattern-classified release errors like the toss, slam, lateral drop, and compressed release, to contact errors including toe contact, heel contact, and outside-of-foot contact, through to grip errors such as hands positioned too high or too low on the ball. More than 40 variations were identified, classified, and fed into the algorithm.
The result for teachers is a model capable of not only grading the skill but providing highly accurate, phase-by-phase diagnostic feedback that tells them exactly what they need to move every student forward.
What do you think? Could you see yourself using this in your school?
www.huddlapp.com.au
