Research map / Multi-agent systems and games / Human-AI and goal-conditioned
The CoachAI Badminton Environment: Bridging the Gap between a Reinforcement Learning Environment and Real-World Badminton Games
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces the CoachAI Badminton Environment, an OpenAI Gym-style RL environment for singles badminton that uses trained ShuttleNet/DyMF/BC opponent AIs and real match data, with visualization and benchmarks.
Problem
Existing RL simulation environments for sports use rule-based opponents or simple physics randomness, and no environment provides authentic opponents for turn-based sports like badminton, causing a gap to real-game deployment.
Approach
Model badminton singles matches as an MDP where the transition dynamics are driven by realistic opponent AIs (ShuttleNet for shot type and landing, DyMF for player movement, and a behavior cloning model for short rallies), trained on ShuttleSet, integrated into a physics-based shuttlecock simulator and extended to full matches following singles rules, with visualization tools.
Key results
- PPO and A2C agents, competing against two opponent AIs over 20 badminton sets, produce differing average score differences, showing algorithm choice matters.
- An RL agent's moving distribution after serves shifted in training, yielding a reported 60% increase in win rate against opponent A.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Shift the PPO agent's serve-return moving distribution from the left court side to the right | PPO agent · Opponent A | win rate: 60% | Helped |
Limitations
- ShuttleNet and DyMF require at least two previous strokes, so a behavior cloning model is needed for early rally states.
- Evaluation is a demonstration with limited benchmark results (two RL algorithms, 20 sets).
Experimental setup
Datasets: ShuttleSet
Benchmarks: badminton singles match score difference (A2C vs PPO against two opponent AIs, 20 sets)
Metrics: average score difference win rate
Baselines: A2C PPO
Builds on: ShuttleNet DyMF OpenAI Gym ShuttleSet
Closest papers by meaning
- The CoachAI Badminton Environment: A Novel Reinforcement Learning Environment with Realistic Opponents (Student Abstract) (AAAI 2024 · similarity 0.96)
- A Reinforcement Learning Badminton Environment for Simulating Player Tactics (Student Abstract) (AAAI 2023 · similarity 0.91)
- Programmatic Modeling and Generation of Real-Time Strategic Soccer Environments for Reinforcement Learning (AAAI 2022 · similarity 0.88)
- Human-Aligned Chess With a Bit of Search (ICLR 2025 · similarity 0.84)
- FightLadder: A Benchmark for Competitive Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.84)
- Designing Skill-Compatible AI: Methodologies and Frameworks in Chess (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.