Back to feed
News Story
APriority81
量子位
1 sources

TennisVAR: AI Coach with Professional Match Data

Researchers from Xiamen University, Shanghai Innovation Institute, and Shanghai Jiao Tong University have introduced TennisVAR, a system that enables AI to perform stroke-evidence-grounded tactical reasoning in tennis matches. They built the large-scale TRACE dataset with 11,189 rally videos and 41,485 stroke events, along with a three-level tactical taxonomy. This work aims to push AI beyond simple action recognition toward understanding tactical logic, offering a new paradigm for sports AI.

SynthePulse Insight · AI deep readingMembers

From "Seeing" to "Understanding": How TennisVAR Makes AI Show Its Reasoning in Tennis

Version 1 · 1 source

When multimodal large models can fluently commentate on matches, do they truly understand tactics? Teams from Xiamen University and others propose TennisVAR and the TRACE benchmark, requiring AI to not only provide conclusions but also locate the shot evidence supporting them, pushing sports AI from "commentary" to "review."

  • TennisVAR proposes a new task: evidence-based tactical reasoning, requiring models to output conclusions while locating key shots and explaining the reasoning process.
  • The TRACE dataset includes 11,189 rally videos, 41,485 shot events, and 25,429 tactical units, covering 109 professional matches and 72 players.
  • In evidence localization, TennisVAR achieves T-F1@8 of 73.04, far surpassing GPT-5.5's 37.03; ablation experiments show significant performance drops when the graph reasoning module is removed.
Open section navigationProblem: AI commentary is fluent, but does it really "understand the game"?

Problem: AI commentary is fluent, but does it really "understand the game"?

Multimodal large models can recognize actions and generate professional commentary, but the research team points out that when a model says "the player wins the point through aggressive play," it may not be able to identify which shots built the advantage or which shot changed the offensive-defensive dynamic. This exposes a deep gap in sports AI: from "seeing actions" to "understanding rules and reviewing decisions."

Traditional methods can accurately detect shot events but often treat each shot independently; video language models can produce seemingly reasonable analyses but may not point to the specific shots supporting their conclusions. The research team summarizes this gap as the "perception-to-understanding" divide.

To fill this gap, research teams from Xiamen University, Shanghai Innovation Institute, and Shanghai Jiao Tong University propose TennisVAR and define a new task—evidence-based tactical reasoning—requiring models to not only identify each shot but also understand why it was played and how multiple shots form tactics.

Free for now

Read the full analysis

4 more sections of analysis, plus the full takeaway

Loading

Credibility boundary

This article's information primarily comes from Qubit's report on the research team's submission, which is a secondary source. All data, model names, and performance metrics are from that report and have not been independently verified. Links to the paper, project page, and code are provided but were not verified in this article.

Sources for this version

  1. 万级回合专业比赛数据，让AI化身你的专属"网球教练"量子位

Primary report

量子位

Primary source