Striatum is predominantly inhibitory and the main input nucleus of the basal ganglia. A functional characterization of its activity dynamics is crucial for understanding the mechanisms underlying phenomenon such as action selection and initiation. Here, we investigated the effects of the spatial connectivity structure on the emergence and maintenance of localized bumps of activity in large-scale striatal networks (10,000 neurons). We show that in striatal network model in which the distance-dependent connection probability varies in a Gaussian fashion (Gaussian networks), the activity remains asynchronous irregular (AI) and spatially homogeneous, independent of the background input. By contrast, when the distance-dependent connectivity varies according to a Gamma distribution (Gamma networks), with short-range connectivity suppressed, a repertoire of activity dynamics can be observed: While weak background inputs induce spatially homogeneous AI activity, stronger background inputs induce stable, spatially localized activity bumps as in 'winner-take-all' (WTA) dynamics. Interestingly, for intermediate background inputs, the networks exhibit spatially localized, but unstable activity bumps (Transition Activity, TA), resembling the experimentally observed neuronal assembly dynamics in the striatum. Among the three main regimes of network activity (AI, WTA, TA) we found that in the AI and TA regimes, network dynamics are flexible and can be easily modified by external stimuli. Moreover, the dynamical state of the network returns to the baseline after the stimulus is removed. By contrast, the dynamics in the WTA state are rigid and can only be changed by very strong external stimuli. These results support the hypothesis that the flexibility of the striatal network state in response to stimuli is important for its normal function and the 'rigid' network states (WTA) correspond to brain disorders such as Parkinson's disease, where the striatum looses its repertoire of dynamic states and is only receptive to very strong inputs.