A typical unmanned ground vehicle (UGV) mission can be composed of various tasks and several alternative paths. Small UGVs are typically teleoperated and rely on electric rechargeable batteries for their operations. Since each battery has limited energy storage capacity, it is essential to predict the expected mission energy requirement during the mission execution and update this prediction adaptively via real-time performance measurements (e.g., vehicle power consumption and velocity). We propose and compare two methods in this paper. One is based on recursive least-squares estimation built upon a UGV longitudinal dynamics model. The other is based on Bayesian estimation when prior knowledge (e.g., road average grade and operator driving style) is available. The proposed Bayesian prediction can effectively combine prior knowledge with real-time performance measurements for adaptively updating the prediction of the mission energy requirement. Our experimental and simulation studies show that the Bayesian approach can yield more accurate predictions even with moderately imprecise prior knowledge. © 2013 Wiley Periodicals, Inc.