Background: Marijuana is among the most widely used controlled substances, with use rising as risk perception declines. Despite growing acceptance, concerns remain regarding its effects on cardiovascular and autonomic physiology, sleep, and risk of cannabis use disorder. Although laboratory and epidemiological studies have characterized these associations, objective evidence from real-world settings is limited. Methods We studied cardiac autonomic function, sleep, and marijuana use patterns in 16,965 consumer-wearable device users, contributing a total of 5,226,739 person-days of data with or without self-reported use of marijuana. Linear mixed models related day-level use to nocturnal cardiac autonomic function (resting heart rate [RHR], heart rate variability [HRV]) and sleep (duration, fragmentation, architecture) at the within- and between-person levels. Binomial mixed models were used to estimate use probability across 72 weeks of marijuana tracking. Results: Members reported marijuana use on a median of 25.1% of logged days (IQR, 7.1-62.6%). Within individuals, use was associated with lower RHR (-0.12bpm; 99.5% CI: [-0.14, -0.10]) and lower HRV (-0.47ms [-0.53, -0.42]). Sleep was longer (+7.60min [7.37, 7.83]) but more fragmented, with more disturbances (+0.47 [0.45, 0.48]), greater wake time (+1.79min [1.70, 1.89]), and altered wearable-estimated sleep architecture. Daily use probability declined over 72 weeks (-11.1% [-11.5, -10.8]). Conclusions: Marijuana use was associated with altered nocturnal cardiac autonomic physiology and longer but more disrupted sleep. Marijuana use declined after commencing tracking, raising the possibility that longitudinal self-monitoring may influence use. These findings extend laboratory evidence on marijuana use to real-world settings and demonstrate the potential of consumer wearables for studying substance use and its physiological correlates at scale.