Split-belt treadmill walking is commonly used for stroke rehabilitation and motor learning experiments. People adjust their gait as they walk on a split-belt treadmill, meaning we can accentuate post-stroke asymmetries to correct those asymmetries overground. We can also use the split-belt treadmill to understand how people adapt to novel motor environments. Researchers use the splitbelt treadmill to explore questions such as how distraction a↵ects motor learning and how reexposure leads to faster readaptation, but we still do not fully understand what drives people’s adaptation. Why do people adjust their step lengths over time when split-belt walking? Why is their adapted gait more energetically ecient than their initial gait? Could people alter their gait to take advantage of the split-belt treadmill? To answer these questions, we considered split-belt walking from the perspective of mechanical energy. We first designed a split-belt rimless wheel simulation model to demonstrate how the split-belt treadmill could act as an external energy source providing power to a person during split-belt walking. We demonstrated that a split-belt rimless wheel can passively walk steadily forward on split belts, even though the same wheel would require energy to walk steadily on belts set to the same speed, known as tied belts. We then conducted a human-subject experiment measuring the energy cost of walking for a variety of both split- and tied-belt speed combinations. We found that the energy cost for human split-belt walking is similar to the cost of tied-belt walking at the average belt speed. Increasing the belt speed di↵erence tends to increase the energy cost of walking. This suggests that people are not able to take advantage of the treadmill power to lower their energy costs. Finally, we conducted an experiment to understand why positive treadmill power during split-belt walking is not metabolically beneficial. We found that people dissipated the treadmill energy rather than using the power e↵ectively, and the gait adjustments required to achieve net positive treadmill work caused other walking costs to increase. Our energy-cost findings inform future split-belt treadmill experiments by enabling researchers to design protocols with consistent levels of energetic load. Our understanding of how people interact with positive treadmill work enables assistive device designers to analyze whether their devices provide work such that people use the work e↵ectively rather than dissipating it.