The mot framework is a system for learning behaviors while organizing them across a two-dimensional, topological map such that similar behaviors are represented in nearby regions of the map. The current paper introduces temporal coherence into the framework, whereby temporally extended behaviors are more likely to be represented within a small, local region of the map. In previous work, the regions of the map represented arbitrary parts of a single global policy. This paper introduces and examines several different methods for achieving temporal coherence, each applying updates to the map using both spatial and temporal neighborhoods, thus encouraging parts of the policy that commonly occur together in time to reside within a common region. These methods are analyzed experimentally in a setting modeled after a human behavior-switching game, in which players are rewarded for producing a series of short but specific behavior sequences. The new methods achieve varying degrees—in some cases high degrees—of temporal coherence. An important byproduct of these methods is the automatic decomposition of behavior sequences into cohesive groupings, each represented individually in local regions.