Research map / Large language models / AI and society
MyoChallenge 2025: A New Benchmark for Human Athletic Intelligence
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.
Reports the MyoChallenge 2025 NeurIPS competition: two musculoskeletal control tracks (table tennis with MyoArm/MyoTorso, soccer penalty kick with MyoLeg/MyoTorso), with winning solutions and benchmark results.
Problem
Lack of a public benchmark integrating high-fidelity full-body musculoskeletal models with interactive physically simulated sports tasks and modern ML control methods.
Approach
A NeurIPS competition built on MyoSuite/MuJoCo that standardizes two athletic tasks (table tennis rally and soccer penalty kick) with phased environment variations, effort-based tie-breaking, and open evaluation on EvalAI; winning solutions combine physics-based planners, muscle synergies, CrossQ, and KINESIS-based transfer with on-policy behavior cloning.
Key results
- Table tennis winner ActingAI reached 100% success in Phase 1 and 64% in Phase 2 with effort 0.1306.
- Soccer winner Servette MyoClub reached 100% success in Phase 1 and 32% in Phase 2, the only team beating the soccer baseline.
- The competition drew 69 teams and 562 submissions.
Surprising findings
- Only one soccer team surpassed the provided baselines, showing the full-body soccer task is very hard.
- Winning soccer policy needed about 14 days of CPU-bound training and table tennis winner 51 hours, motivating GPU-accelerated simulation.
Limitations
- Soccer task proved too hard, with few teams beating baseline.
- CPU-bound simulation makes training of full-body models take weeks.
- No participants from South America, Oceania, or Africa; limited diversity.
Experimental setup
Datasets: KIT-Locomotion AMASS BONES-SEED
Benchmarks: MyoChallenge 2025 table tennis track MyoChallenge 2025 soccer track
Metrics: success rate (%) muscle activation effort
Baselines: PPO baseline (table tennis) DEP-RL reflex-based controller
Builds on: MyoSuite KINESIS CrossQ PPO
Closest papers by meaning
- Muscles in Time: Learning to Understand Human Motion In-Depth by Simulating Muscle Activations (NeurIPS 2024 · similarity 0.84)
- Motion Control of High-Dimensional Musculoskeletal Systems with Hierarchical Model-Based Planning (ICLR 2025 · similarity 0.82)
- MS-MANO: Enabling Hand Pose Tracking with Biomechanical Constraints (CVPR 2024 · similarity 0.82)
- EMGBench: Benchmarking Out-of-Distribution Generalization and Adaptation for Electromyography (NeurIPS 2024 · similarity 0.82)
- SportR: A Benchmark for Multimodal Large Language Model Reasoning in Sports (ICLR 2026 · similarity 0.82)
- emg2qwerty: A Large Dataset with Baselines for Touch Typing using Surface Electromyography (NeurIPS 2024 · similarity 0.81)
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.