VolleyAI

About VolleyAI

Science-backed strategy for every rally.

VolleyAI is a volleyball tactical decision-support system built on geometric deep learning. We adapt the graph-neural-network paradigm — proven by Google DeepMind’s TacticAI for football corner kicks (Liverpool FC, 2024) — to volleyball’s structured phases: serve receive, blocking, setter decision-making, and defensive positioning.

The research

Our reference implementation, documented in VolleyAI: A Graph Neural Network Assistant for Volleyball Tactical Decision Support, formulates three core tasks: (1) receiver prediction in serve receive, (2) attack-outcome prediction conditioned on setter decisions, and (3) guided generative refinement of player formations to maximize rally win probability.

  • • 81.2% rally outcome accuracy (Graph Transformer)
  • • 59.1% set-location accuracy (GCN)
  • • 86.4% hit-type accuracy
  • • Generative suggestions rated indistinguishable from real formations 72% of the time, preferred 68% in blind coach comparisons

Preprint submitted to Nature Communications / MIT Sloan Sports Analytics Conference 2026. Code: github.com/volleyai/volleyai (MIT). Data: VREN (CC-BY-4.0).

Equivariant by design

VolleyGraph respects court rotations via D₂ × C₆ symmetry — reads hold for any rotation or reflection.

Built on real data

Trained on the VREN dataset (7,000+ NCAA/professional rallies) and multi-camera tracking.

Human-in-the-loop

Validated with NCAA Division I coaches — suggestions preferred over existing tactics in blind tests.

Coach with the math on your side.