Predicting Mouse Positions Beyond a System's Latency Can Increase Throughput and User Experience in Linear Steering Tasks

Latency is present in all interactive systems and decreases user experience and performance. Previous work developed approaches that predict user actions and show these predictions to reduce latencies' negative effects. While this can increase user experience and performance, it is unclear if predicting beyond a system's latency results in further improvements. Therefore, we investigated the effects of predicting beyond a system's latency. We collected data from 60 participants performing Steering Law tasks to systematically train an artificial neural network (ANN) that predicts 100ms into the future. We integrated the ANN into the Steering Law task and buffered users' inputs to simulate latency between 50ms and -50ms. A study with 30 participants showed that decreasing latency beyond the system's latency increases throughput up to -50ms. Subjective measures improved up to -16.67ms without negative effects on agency. Overall, we show that predicting beyond a system's latency can increase performance and user experience.