DShuttleNet: A Badminton Doubles Dataset for Evaluating Team Coordination

arXiv:2610.14016 · cs.LG · Submitted 2026-09-29 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.

Tom: Today's paper: "DShuttleNet: A Badminton Doubles Dataset for Evaluating Team Coordination".

Jane: The gist The authors present DShuttleNet, a comprehensive dataset for badminton doubles curated from elite BWF World Tour tournaments,

Tom: First, who's behind it and why it matters.

Title and authors: Tom: So we’re looking at this paper, "DShuttleNet: A Badminton Doubles Dataset for Evaluating Team Coordination." It’s about how we can finally get data that actually looks at what's happening between the teammates in a badminton doubles match.

Jane: Exactly. Before this, most datasets were focused on individual players or just the ball itself, which means they totally missed how much coordination matters in doubles, right?

Tom: Right. This paper presents DShuttleNet as a comprehensive dataset for badminton doubles that includes expert-labeled "Loss Reasons." It’s designed to be a diagnostic tool for causal tactical evaluation.

Lu: What’s really interesting is that they're the first public stroke-level badminton doubles dataset with synchronized spatial positions for all four players. They have fifty-eight matches, four thousand four hundred eighty-three rallies, and fifty-four thousand one hundred forty-one shots from elite BWF World Tour tournaments between the years two thousand twenty-two and two thousand twenty-five <ref:2610.14016#pg1,from elite BWF World Tour>.

Meng: fifty-four thousand shots is a lot of data to wrangle. What’s the main thing they’re doing differently compared to what we have now?

Tom: They are adding this expert labeling of "Loss Reasons," which helps them diagnose tactical failures beyond just who hit the ball or where it went. They say this helps us understand the causal flow of a rally because they've annotated these shots shot-by-shot using domain experts.

Jane: That sounds really detailed. So, how do they represent that movement spatially? Because if you’re talking about coordination, you need to see the angles between players.

Tom: They formally define two geometry-driven representations called Teammate-Relative Angle, or TRA, and Vertical Shot Angle, or VSA. The TRA is defined as the minimum angle between the line connecting two teammates and a line parallel to the net; lower angles mean a side-by-side defensive formation.

Lu: And VSA captures verticality—it distinguishes between "Downward," "Flat," and "Upward shots"—which they say is critical for bridging raw movement with high-level tactical transitions. This helps models understand the vertical intent behind each stroke.

Meng: So, they're not just tracking where a player was; they’re tracking the geometry of the team structure itself, like how tight or spread out they are in real time.

Jane: That makes sense because standard models often underfit when they don't incorporate these coordination features. They introduce two novel metrics to quantify synergy: the Court Coverage Efficacy Index, or CCEI, and the Threat Load Asymmetry Index, or TLAI.

Title and authors: Tom: CCEI measures the success rate of the non-hitting partner in executing reciprocal compensation by checking if at least one player is within a radial threshold R around a central pivot point C. It checks for that necessary back-and-forth adjustment between players.

Lu: And TLAI quantifies lateral spatial coordination by modeling defensive responsibility zones to calculate TLAI equals the absolute difference between two threat load zones divided by their sum. It’s a way to assess the imbalance of the threat load between defenders.

Meng: So, if you can measure that asymmetry, you can see exactly where communication or coverage breaks down spatially on the court.

Jane: And they show some empirical validation on this stuff. They found that ignoring these coordination-centric features leads to underfitting in conventional models. For instance, when predicting behavior, ignoring the TRA results in a Formation MSE of ten to fifteen degrees <ref:2610.14016#pg1>.

Tom: And their CCEI transition probabilities show something specific: with Coverage equal to one—that’s CCEI equals one—you get a sixty-six point eight percent attack continuation rate, whereas when Coverage is zero, that rate drops to fifty-three point nine percent.

Lu: That suggests the formation is directly linked to offensive success in this way. They also found that TLAI reveals a critical threshold at TLAI equal to approximately zero point two where the failure rate escalates sharply, peaking at twenty-two point five percent in extreme scenarios when TLAI equals zero point four.

Meng: That peak at zero point two and zero point four sounds like a really clear indicator for someone trying to optimize defensive positioning in real time.

Jane: What this means for someone just listening is that if you’re looking at doubles play, you can finally quantify when the team structure itself is failing them, not just when one player misses a shot.

Tom: They also mention several downstream tasks they think are really important: Coordination-aware Behavior Prediction, Action-Value Estimation, Defensive Coverage and Failure Analysis, and Match Intensity and Fatigue Modeling.

Lu: For behavior prediction using DShuttleNet’s data, if you ignore the TRA angle, your trajectory predictions get oversmoothed with that ten to fifteen degree Formation MSE we just discussed <ref:2610.14016#pg1>.

Meng: If you’re deploying a model, knowing that ignoring the TRA leads to a poor formation prediction is something I can use. It tells me where my current training data is missing crucial spatial dynamics.

Title and authors: Jane: And Action-Value Estimation, or AVE, they propose using the TRA and CCEI to determine if a specific stroke successfully facilitates an offensive transition or forces the team into a vulnerable defensive state.

Tom: That means you value a shot based not just on its immediate outcome, but on its capacity to optimize your partner’s subsequent spatial options and the team’s overall win probability.

Lu: They also have this defensive coverage analysis system that quantifies the integrity of defensive structures using CCEI and TLAI to visualize miscommunication zones where neither partner effectively covers an incoming threat.

Meng: Visualizing those specific miscommunication zones based on TLAI is exactly what I need when I'm looking at production data and trying to spot a systemic failure in team coverage.

Jane: And they suggest a fatigue modeling system that tracks how exhaustion degrades team synergy by correlating it with metrics, specifically observing how fatigue reduces the volatility of the TRA and correlates with an escalation in TLAI.

Tom: So, if you’re looking at stamina management for a doubles team, this gives you data-driven insights to optimize that strategy.

Lu: This work is really important because it shows that focusing on these coordination-centric features isn't just academic; conventional models underfit when they fail to incorporate them, which justifies the need for DShuttleNet.

Meng: I see how this could help with deployment stability; knowing what breaks at scale based on spatial asymmetry metrics is valuable.

Jane: So, in summary, DShuttleNet provides a way to move beyond ball-centric data and look at the actual team dynamics of badminton doubles through geometry and coordination metrics.

Tom: Yeah, it gives us the tools to build systems that predict behavior and value shots based on team synergy rather than just isolated actions.

Lu: This paper opens up new research directions for analyzing how these features affect prediction models, which is a big step in representation learning for sports data.

Meng: For me, the practical implication is having a better diagnostic tool to find exactly where our current AI systems are failing to capture the complexity of doubles play.

Jane: So, we’ve seen how they built these metrics and what their initial findings show about formation and threat load asymmetry.

Title and authors: Tom: We’re going to wrap up by summarizing the core contributions of DShuttleNet: creating a dataset with expert loss reasons, defining TRA and VSA geometry, and introducing CCEI and TLAI for synergy quantification.

Lu: The paper emphasizes that these coordination-centric features are critical because conventional models suffer from underfitting when they don't include them.

Meng: It’s about moving the evaluation from just individual performance to how the team works together spatially, which is a necessary step for robust AI modeling in this domain.

Jane: And while they show these metrics are diagnostic, they also point out a limitation: these coordination metrics can't capture things like psychological pressure or deliberate tactical sacrifices.

Tom: Plus, because DShuttleNet is from elite BWF World Tour tournaments, the tactical patterns might not fully generalize to recreational players.

Lu: And the authors recommend that empirical validation on non-elite populations should happen before drawing conclusions about broader player bases.

Meng: So, the limitation is that these metrics are for high-level analysis and not necessarily absolute scoring rubrics for automated refereeing systems right now.

Jane: That’s a good point to keep in mind when thinking about real-world application versus pure research.

Tom: We’ve covered the mechanics of DShuttleNet, from the data collection to the specific metrics they developed like TRA and VSA.

Lu: The paper sets up clear directions for using these features in downstream tasks like coordination-aware behavior prediction and defensive coverage analysis.

Meng: For deployment, this means we can start building systems that actively try to predict when a defensive structure is about to fail based on TLAI values.

Jane: It really shows how incorporating spatial awareness directly into the input features makes a difference in model performance metrics like the Formation MSE and attack continuation rates.

Tom: So, this paper provides a solid foundation for building more sophisticated AI systems that understand the interplay between players rather than just tracking individual trajectories.

Lu: This research is significant because it validates that focusing on team coordination data can lead to better representation learning for complex sports like doubles.

Meng: It’s a big step in making AI models more context-aware when dealing with human interaction in dynamic settings.

Jane: And that’s what we were talking about today: DShuttleNet, the first expert-annotated dataset for causal evaluation in badminton doubles.

The paper's summary: Tom: So, DShuttleNet is essentially presenting the first public dataset for badminton doubles that actually tracks synchronized positions for all four players in every rally, and they’ve added these expert-labeled loss reasons to help diagnose tactical failures.

Jane: So what does that mean for us? It means we can finally move beyond just looking at individual shots or trajectories and start analyzing the entire team structure during a match.

Tom: Exactly. They define these geometry-driven representations, like Teammate-Relative Angle, or TRA, which tells you about the formation—like if everyone’s side-by-side defensively. And they have Vertical Shot Angle, or VSA, which is about verticality—downward versus upward shots—to see the intent behind each stroke.

Lu: From a representation learning standpoint, it's interesting how they tie those angles to real tactical dynamics; the TRA seems to directly map onto offensive versus defensive alignments.

Meng: For me, what’s cool is that they introduce these coordination metrics like Court Coverage Efficacy Index, or CCEI and Threat Load Asymmetry Index. They’re trying to quantify synergy and spatial imbalance.

Jane: The CCEI measures if the non-hitting partner is compensating effectively, which gives us a concrete way to measure that reciprocal effort on the court. And the TLAI quantifies how unevenly the defensive responsibility is distributed between players.

Tom: That TLAI calculation, it’s really useful because they found a critical threshold around TLAI equal to zero point two where failure rates start escalating sharply, peaking at twenty-two point five percent when it hits zero point four.

Meng: If I’m running a model to predict outcomes, seeing that the risk of losing a point jumps so fast after that imbalance level makes sense for production stability.

Jane: So, what does this change for someone just listening? It changes how you think about doubles play. You’re not just watching who hits the hardest; you’re seeing the geometry of the team and whether they are effectively compensating each other when things get tough.

Tom: Right. Conventional models, as Lu mentioned, underfit when they ignore these coordination features; ignoring the TRA leads to trajectory predictions with a formation mean square error of ten to fifteen degrees.

Lu: And if you look at their findings on CCEI, that one transition probability—where coverage is one yielding a sixty-six point eight percent attack continuation rate versus fifty-three point nine percent when coverage is zero—it shows the direct causal link between proper team positioning and offensive success.

Jane: It’s really concrete. So, the paper argues that incorporating these spatial features isn't just academic; it directly improves the performance of predictive models by giving them context about how the team is actually set up.

Tom: And they suggest several downstream applications, like coordination-aware behavior prediction and defensive coverage analysis, which are really practical things for building better AI systems for this sport.

Lu: The implication is that we can build systems that don't just track individual movements but actively predict when a team structure is about to break down based on those spatial metrics.

Meng: That gives me a clear direction for what I would actually deploy, focusing on identifying those miscommunication zones visualized by the TLAI metric.

Jane: So, DShuttleNet provides the data and the mathematical framework to build AI that understands team coordination in badminton doubles at a level that goes beyond individual performance metrics.

Tom: We’ve seen how they built these metrics and what their initial findings show about formation and threat load asymmetry, which sets up a lot of new research directions for representation learning in sports data.

The paper's improvements: Tom: So, DShuttleNet isn't just presenting data; they’re suggesting how to actually *use* it to build better AI tools for sports analysis, and they propose four specific areas where these metrics can help.

Jane: What are these improvements? Because we talked about the metrics like CCEI and TLAI, what do they suggest we should be doing with that information?

Tom: First, Coordination-aware Behavior Prediction. Instead of just predicting where one player will go, you get to forecast future trajectories, angular configurations, and shot type probabilities by explicitly incorporating teammate interactions.

Lu: That addresses the issue we found earlier where ignoring the TRA leads to oversmoothed trajectory predictions with a formation mean square error of ten to fifteen degrees.

Meng: For production deployment, that means we can build systems that predict future movement based on what your partner is doing, not just isolated shots. It makes the prediction much more robust for real-time use.

Jane: Then there’s Action-Value Estimation, or AVE. They propose using the TRA and CCEI to determine if a specific stroke actually sets up an offensive transition or pushes the team into a vulnerable defensive state.

Tom: That means you value a shot based on its capacity to optimize your partner’s subsequent spatial options and the team’s overall win probability, not just by looking at what happens immediately after it.

Lu: It moves the valuation from an immediate outcome to a synergy-centric framework where the shot's merit is judged by its impact on future spatial options.

Jane: Next, they suggest a defensive coverage analysis system. This uses CCEI and TLAI to quantify the integrity of defensive structures, allowing you to visualize miscommunication zones where neither partner is covering an incoming threat effectively.

Meng: Visualizing those specific failure points based on TLAI is exactly what I need when I’m looking at production data to spot systemic failures in team coverage.

Tom: Finally, they propose a fatigue modeling system that tracks how exhaustion degrades team synergy by correlating it with metrics, specifically observing how fatigue reduces the volatility of the TRA and correlates with an escalation in TLAI.

Lu: That correlation between fatigue and the TLAI escalation gives us data-driven insights to optimize stamina management strategies for athletes.

Jane: It seems like this paper is moving us toward building AI that understands not just individual skill, but how those skills translate into team dynamics under pressure and exhaustion.

Tom: So, DShuttleNet isn't just a dataset; it’s a toolkit for building systems that value the whole team performance based on spatial coordination metrics.

Meng: It’s about moving from tracking isolated actions to modeling the complex interplay between players in dynamic settings, which is crucial for any robust AI application.

Conclusion: Tom: So we’re wrapping up on DShuttleNet, which is this dataset for badminton doubles that introduces expert labeling for loss reasons to help us diagnose tactical failures by looking at team coordination metrics like CCEI and TLAI.

Jane: Exactly. This paper is about moving beyond just individual shots and starting to understand the team dynamics in a rally.

Lu: The core idea is that conventional models underfit when they don't incorporate these geometry-driven representations, like TRA and VSA, because ignoring them leads to poor trajectory predictions with a mean square error of ten to fifteen degrees.

Meng: For me at the startup level, the real value here is seeing how we can build better behavior prediction systems that actually account for the formation dynamics instead of just guessing based on isolated player data.

Lalam: I’m fairly confident in this direction; incorporating those spatial features directly into representation learning makes a lot more sense for capturing complex human interaction.

Tom: And they show that incorporating these metrics into Action-Value Estimation can really change how we value a stroke, making it based on its ability to set up the partner rather than just the immediate point outcome.

Jane: It means we get a clearer picture of what makes a good shot not just in terms of physics, but in terms of team synergy and tactical positioning.

Lu: And their fatigue modeling system is pretty compelling because it shows that exhaustion directly degrades those coordination metrics, specifically reducing the volatility of the TRA and increasing the TLAI.

Meng: That’s critical for deploying anything in a real competition setting; knowing how fatigue impacts these specific spatial metrics gives us actionable data for managing player stamina.

Tom: So, DShuttleNet provides this comprehensive data set with expert annotations and new metrics that allow us to build AI tools capable of predicting behavior and valuing actions based on team synergy.

Jane: It’s a big step in making AI models more context-aware when dealing with human interaction in dynamic sports settings.

Lu: This research is significant because it validates that focusing on team coordination data can lead to better representation learning for complex sports like doubles.

Meng: It’s a big step in making AI models more context-aware when dealing with human interaction in dynamic settings, which is what I need to see for production stability.

Lalam: DShuttleNet offers a solid foundation for building more sophisticated AI systems that understand the interplay between players rather than just tracking individual trajectories.

Guan-Yi Jhang, Chiencheng Chiang, Wei-Kai Chen, Hui-Ching Yu, Hsu-Chao Lai, Hwai-Jung Hsu, Wang Chih-Chuan, Jhy-Cheng Huang, Jiun-Long Huang

National Yang Ming Chiao Tung University · Chang Gung University · Feng Chia University

cs.LG

Submitted: 2026-09-29

Updated: 2026-09-29

Code: https://github.com/xgabora/Club-Football-Match-Data-2000-2025

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 86/100

The gist: The gist The authors present DShuttleNet, a comprehensive dataset for badminton doubles curated from elite BWF World Tour tournaments, which uniquely incorporates expert-labeled "Loss Reasons" to

Key concepts

Teammate-Relative Angle (TRA): This measures the minimum angle between two teammates' lines relative to the net. A lower angle suggests a side-by-side defensive setup, while a higher angle indicates an offensive formation where players are positioned front and back.
Vertical Shot Angle (VSA): VSA distinguishes shot verticality into 'Downward,' 'Flat,' and 'Upward' categories. This feature helps models understand the vertical intent behind a stroke, bridging raw movement data with high-level tactical decisions in doubles.
Court Coverage Efficacy Index (CCEI): CCEI quantifies how well the non-hitting partner compensates for an opponent's action. It checks if at least one teammate is within a specific distance of a central pivot point, measuring the success rate of reciprocal defensive positioning.
Threat Load Asymmetry Index (TLAI): TLAI measures lateral spatial coordination by modeling defensive responsibility zones. A high TLAI indicates an imbalance in threat load between players, and the analysis shows that when this index is too high, the failure rate of the team increases sharply.

Summary

The gist The authors present DShuttleNet, a comprehensive dataset for badminton doubles curated from elite BWF World Tour tournaments, which uniquely incorporates expert-labeled Loss Reasons to serve as a diagnostic tool for causal tactical evaluation.

Dataset Construction and Annotation

DShuttleNet is the first public stroke-level badminton doubles dataset with synchronized spatial positions for all four players, comprising 58 matches, 4,483 rallies, and 54,141 shots from top-tier international tournaments between 2022 and 2025 The annotation process involves three main stages: Match Video Preparation to standardize the annotation, Match Video Preprocessing to streamline the pipeline, and Shot-by-shot Labeling Domain experts annotated the preprocessed clips chronologically using a single-pass workflow To guarantee professional-grade accuracy, domain experts randomly sampled and reviewed annotated rallies to verify the accuracy of spatial trajectories, scoring logic, and stroke-level records

Geometry-driven Representations

The dataset formally defines two geometry-driven representations to characterize coordination and shot aggressiveness: Teammate-Relative Angle (TRA) and Vertical Shot Angle (VSA) The TRA is defined as the minimum angle between the line connecting two teammates and the line parallel to the net, with lower angles indicating a side-by-side defensive formation and higher angles representing a front-back offensive alignment The VSA captures verticality, distinguishing between Downward, Flat, and Upward shots, which serves as a critical bridge between raw movement and high-level tactical transitions These features allow models to understand the causal flow of a rally by encoding the vertical intent behind each stroke

Coordination Metrics

To quantify interpersonal synergy, DShuttleNet introduces two novel metrics: Court Coverage Efficacy Index (CCEI) and Threat Load Asymmetry Index (TLAI) The CCEI measures the success rate of the non-hitting partner in executing reciprocal compensation by checking if at least one player is positioned within a radial threshold R around a central pivot point C The TLAI quantifies lateral spatial coordination by modeling defensive responsibility zones and calculating T LAI = T L1 − T L2 / (T L1 + T L2) to assess the imbalance of the threat load between defenders Empirical validation demonstrates that conventional models suffer from underfitting when failing to incorporate these coordination-centric features, highlighting their critical importance

Analytical Applications and Future Directions

The proposed dataset supports several downstream tasks, including Coordination-aware Behavior Prediction, Action-Value Estimation (AVE), Defensive Coverage and Failure Analysis, and Match Intensity and Fatigue Modeling For behavior prediction, ignoring the TRA leads to oversmoothed trajectory predictions with a Formation MSE of 10–15 degrees The CCEI transition probabilities show that With Coverage(CCEI = 1) yields a 66.8% attack continuation rate, whereas No Coverage(CCEI = 0) reduces this to 53.9% Furthermore, the TLAI analysis reveals that a critical threshold emerges at TLAI ≈ 0.2 where the failure rate escalates sharply, peaking at 22.5% in extreme scenarios (TLAI= 0.4)

Limitations and Implications

The coordination metrics, CCEI and TLAI, are designed as diagnostic instruments for causal tactical evaluation but cannot capture unobservable factors such as psychological pressure or deliberate tactical sacrifices Furthermore, because DShuttleNet is curated entirely from elite BWF World Tour tournaments, the captured tactical patterns may not fully generalize to recreational or amateur players Responsible use requires awareness that these metrics cannot be applied as absolute scoring rubrics in automated refereeing systems Finally, empirical validation on non-elite populations is recommended before drawing cross-level tactical conclusions

The paper presents DShuttleNet, the first expert-annotated dataset for causal evaluation in badminton doubles.

How it works

Constructing a High-Quality, Expert-Annotated Dataset for Causal Evaluation: The authors present the first public stroke-level badminton doubles dataset, comprising 58 matches, 4,483 rallies, and 54,141 shots from top-tier international tournaments between 2022 and 2025 Unlike existing datasets that focus on individual trajectories, DShuttleNet provides synchronized spatial positions for all four players and uniquely incorporates expert-annotated Loss Reasons categorized into six mutually exclusive types
Defining Geometry-driven and Formation-aware Representations: The authors formally define two geometric metrics, Teammate-Relative Angle (TRA) and Vertical Shot Angle (VSA) to characterize team coordination and shot aggressiveness in the context of doubles By integrating both angular features, they establish a data-driven framework that provides a deep understanding between team formations and tactical dynamics
Proposing a Quantitative Metric for Teammate Coordination: The authors introduce two novel metrics, Court Coverage Efficacy Index (CCEI) and Threat Load Asymmetry Index (TLAI) to address the challenge of quantifying interpersonal synergy CCEI measures the success rate of the non-hitting partner in executing reciprocal compensation, while TLAI quantifies lateral spatial coordination by modeling defensive responsibility zones

Key Findings and Contributions

The experimental results demonstrate that conventional models suffer from underfitting when failing to incorporate these coordination-centric features, highlighting the critical importance and potential impact of releasing DShuttleNet The analysis reveals a distinct correlation between VSA and team formations, where offensive downward shots concentrate at high-angle intervals, typically triggering a front-and-back attacking formation The study also confirms that the risk of point loss accelerates as spatial imbalance worsens under TLAI, with failure rates peaking at 22.5% when TLAI= 0.4

New Research Directions

The proposed DShuttleNet supports several downstream tasks that capture the unique tactical dynamics of professional badminton doubles, including Coordination-aware Behavior Prediction, Action-Value Estimation (AVE), Defensive Coverage and Failure Analysis, and Match Intensity and Fatigue Modeling These capabilities have the potential to benefit professional coaches and analysts by allowing them to systematically identify defensive vulnerabilities and optimize stamina management strategies

The authors present DShuttleNet, the first expert-annotated dataset for causal evaluation in badminton doubles.

References

[1] E. Collet Shuttlecock velocity decay after smash and slice shots in badminton 101(12):125007 mar 2026 doi: 10.1088/1402-4896/ae5361 URL https://doi.org/

[2] A Deliege, A Cioppa, S Giancola, M J Seikavandi, J V Dueholm, K Nasrollahi, B Ghanem T B Moeslund and M Van Droogenbroeck Soccernet-v2: A dataset and benchmarks for holistic understanding of broadcast soccer videos In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition pages 4508–4519 2021 doi: 10.1109/CVPR.

[3] N Ding, K Takeda, W Jin, Y Bei and K Fujii Estimation of control area in badminton doubles with pose information from top and back view drone videos Multimedia Tools and Applications 83(8):24777–24793 Aug.

Improvements for AI systems

A coordination-aware behavior prediction system can be developed to forecast future trajectory, angular configuration, and shot type probabilities by explicitly incorporating teammate interactions rather than relying on isolated trajectories. This addresses the finding that ignoring the TRA leads to oversmoothed trajectory predictions, which fail to capture the rapid rotations and sharp directional shifts inherent in professional doubles.
An Action-Value Estimation (AVE) framework can be implemented that determines whether a specific stroke successfully facilitates an offensive transition (front-back formation) or forces the team into a vulnerable defensive state by utilizing the TRA and CCEI. This enables a synergy-centric valuation framework, where a shot’s merit is adjudicated not only by its immediate outcome but by its capacity to optimize the partner’s subsequent spatial options and the team’s overall win probability.
A defensive coverage analysis system can be built that quantifies the integrity of defensive structures using CCEI and TLAI to visualize miscommunication zones where neither partner effectively covers the incoming threat. This allows for a data-driven methodology for partner-specific performance profiling that accounts for the essential coordination in doubles.
A fatigue modeling system can be introduced that tracks how exhaustion degrades team synergy by correlating it with metrics, specifically observing how fatigue reduces the volatility of the TRA and correlates with an escalation in the TLAI, providing data-driven insights to optimize stamina management.

Abstract

Most existing sports datasets are individual-centric or ball-centric, focusing on individual player actions while ignoring interactions among teammates. In badminton doubles, tactical success is governed by rapid team formation and coordination, yet standard resources for modeling these player interactions are scarce. To bridge this gap, we present DShuttleNet, a comprehensive dataset for badminton doubles curated from elite BWF World Tour tournaments. Beyond trajectory data, DShuttleNet uniquely incorporates expert-labeled "Loss Reasons," transforming the dataset into a diagnostic tool for causal tactical evaluation. We formally define two geometry-driven representations, Teammate-Relative Angle (TRA) and Vertical Shot Angle (VSA), to characterize coordination and shot aggressiveness. To quantify interpersonal synergy, we introduce novel metrics, Court Coverage Efficacy Index (CCEI) and Threat Load Asymmetry Index (TLAI), which evaluate the often-overlooked non-hitting partner's role and defensive coordination. Experimental results demonstrate that conventional models suffer from underfitting when failing to incorporate these coordination-centric features, highlighting the critical importance and potential impact of releasing DShuttleNet. Dataset is available at: https://doi.org/10.34740/kaggle/dsv/16097938.

Sources

BFMD: A Full-Match Badminton Dense Dataset for Dense Shot Captioning