Answers are grounded in this paper and cite their sources.
Conventional college fitness training in China relies on uniform schedules that ignore between-student variability, producing marginal gains for many and elevated risk for those pushed beyond their adaptive ceiling. We propose an adaptive reinforcement learning framework that generates and revises individualized training plans aligned with the National Student Physical Health Standard. The framework binds five layers—input ingestion, adaptive feature extraction, an improved Proximal Policy Optimization decision module, plan rendering, and dual-rate feedback—into a closed loop that ingests fitness scores, demographic descriptors, and longitudinal training logs to issue weekly prescriptions across endurance, strength, and flexibility components. Because no agent may experiment on students, we treat the historical records as an offline decision problem: a validated student-response model supplies the state transitions, and returns for prescriptions the records never contain are estimated only inside the support of the teaching policy that generated them. Three design elements distinguish the system: a slowly refreshed adaptive state representation that absorbs cohort drift; a four-component reward with deficit-driven scalarization balancing fitness gain, safety, adherence, and plan rationality; and a bounded-staleness prioritized buffer with importance-sampling correction, paired with cosine-annealed learning rates, for stable convergence under noisy educational signals. Experiments on 4128 students from four cohorts, partitioned so that no student ever spans two folds, returned a composite fitness improvement of 14.6% in held-out cross-validation. A 16-week randomized field study with 312 participants produced a 13.8% gain against 7.1% in the control arm; the prespecified intention-to-treat mixed model places the adjusted between-arm effect at 4.71 composite points (95% CI 3.92–5.50), or roughly six percentage points of relative improvement. Plan rationality reached 91.7% and adherence 87.3%, with the largest absolute gains accruing to initially failing students. Adaptive reinforcement learning therefore offers a workable—though not yet longitudinally proven—path toward precision physical education in higher education settings.
No takes yet. Share an insight, caveat, or question.
Zhou et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: