It is commonly assumed that path integration is based on an extrinsic measure of the objective distance traversed during locomotion. In contrast, biological odometers may rely on embodied intrinsic measures, such as idiothetic information specific to an action mode. We investigated this question using a distance reproduction task in which participants traveled an outbound distance and then reproduced that distance using the same or a different action mode. The extrinsic model predicted that distance reproduction should be invariant across action modes, whereas the intrinsic model predicted invariance only within an action mode. In Experiment 1, we held the outbound mode constant while varying the response mode (walk–walk, walk–throw) and corrected for response production error (view–walk, view–throw). In Experiment 2, we crossed different gaits in the outbound and response modes (walk, gallop). In both cases, we found that distance reproduction was significantly more accurate when the outbound and response modes matched, consistent with the intrinsic model. The results indicate that the human odometer preferentially relies on an intrinsic, rather than an extrinsic, metric. This solution is sufficient to support successful path integration within an action mode (but not across action modes), without the difficulties of objective distance estimation.