Fig. 2: Evaluating OpenMetabolics’ real-world energy expenditure estimation using a smartphone strapped to the thigh.

a OpenMetabolics was compared to a commercial smartwatch and portable respirometry system. b Participants performed physical activity experiments on public sidewalks. c Map of the 650-m course used for evaluation. Participants performed 20 bouts of walking through the course while following ecologically relevant⁴⁸ audio prompts⁴⁹ to elicit naturalistic walking. Each lap included two flights of stair climbing. Distribution of self-selected walking speeds (d) and walking bout durations (e) during the experiment, compared with previously recorded distributions of real-world walking data¹⁵. f OpenMetabolics had the lowest weighted absolute error across all physical activities compared to existing physical activity monitors. g OpenMetabolics had significantly lower cumulative energy expenditure error than a commercial smartwatch, heart rate model, pedometer, and thigh-based accelerometer during real-world walking, compared to respirometry (multiple comparison tests, n = 28, *P < 0.05, **P < 0.005). h OpenMetabolics had lower absolute error in cumulative energy expenditure across several physical activities (multiple comparison tests, n = 10, *P < 0.05, **P < 0.005). The error for each activity was weighted based on the number of participants. The boxes show the interquartile range, with a line at the median and a dot and percentage value representing the mean. Whiskers extend to 1.5 times the interquartile range.