Tensegrity structures have non-conventional structural properties since they require self-stress in order to be stable. When a tensegrity structure integrates sensors, actuators, (structural members able to change their length) and a control system, deployment and modification of structural behavior becomes feasible. Control of tensegrity structures is a special type of control since multiple-degrees of freedom depend on a lower number of coupled actuators. In addition, there is no direct closed-form solution for achieving a given control task, and this complicates the search for control commands. This thesis investigates the behavior of a near full-scale deployable tensegrity footbridge that deploys from both sides and connects at mid-span. Tests on the structure demonstrate that deployment behavior is not reproducible and therefore, pre-defined control commands are unsuccessful for achieving desired positions. Active control is thus, required for mid-span connection of the two bridge halves. Available simulation models involve assumptions that are not satisfied on the near full-scale structure. In this case non-dimensional, friction free joints are poor assumptions. Using measurements of the response in real-time integrates variable behavior. This advantage shows potential for controlling full-scale structures where changing environment conditions are likely. The active-control methodology for the mid-span connection of the two bridge halves, combining simplified simulations and real-time measurements, accommodates lack of accuracy due to modelling simplifications and non-repeatable behavior. A strategy to partially re-use previous control command reduces the time required for the mid-span connection. In this way, the structure learns from its experience. A damage detection and location method that includes target reliability is tested for single cable damage situations. Despite important modeling simplifications, it is possible to focus on possible damage locations through combination of actuation, measurements, simulations and data interpretation techniques. The structure thus performs self-diagnosis. Active control for structural behavior enhancement and adaptation to damage (self-adaptation) is successfully tested. Simulations are used in order to avoid testing control command that lead to excessive internal forces. Real-time measurements provide a direct evaluation of the actuation effect. Stochastic search is the component that guides the interaction with measurements. Combination of these components results in adaptation and learning since the control system gradually improves performance of the structure through the interaction with the environment during the stochastic search process. Self-diagnosis, self-adaptation and learning are features that are usually associated with living organisms. Thus, this research contributes towards behavior-based biomimetics in structural engineering.