LiftSafe: Predicting Unsafe Repetitions in Strength Training using a Single Wearable Sensor

Strength training is essential for improving muscle
strength and overall fitness, but it also carries the risk of
fatigue-driven injury. Existing sensing systems provide reactive
feedback, analyzing form or fatigue only after a repetition is
completed, which limits their ability to prevent unsafe events.
In this paper, we present LiftSafe, a monitoring system that
proactively predicts whether the next repetition in a strength
training session will be safe or unsafe using only a single low-cost
inertial measurement unit on the upper arm. LiftSafe employs a
time-series forecasting model to predict the movement of the
upcoming repetition, and uses contrastive learning to derive
embeddings that capture subtle fatigue-related characteristics
beyond kinematic features. A progression-aware fusion module
then integrates these embeddings from forecasted and recently
completed repetitions to estimate risk for the next repetition.
To demonstrate LiftSafe’s feasibility, robustness, and practicality,
we showcase its performance for 24 participants following an nRep Max study protocol for bench press, overhead press, and
lateral press exercises. LiftSafe achieved a balanced accuracy of
87% under leave-one-subject-out validation and demonstrated
generalization across exercises and subjects. By transforming
exercise monitoring from reactive assessment to proactive safety
prediction, LiftSafe paves the way for safer and accessible
strength training.