Autonomic Computing and self-adaptive systems are a response to the increasing complexity required to cope with changing environments and varying system resources. However, the complexity of the adaptation logic itself increases with the available information in particular for distributed systems. This leads to uncertainty at runtime resulting in incompleteness in the representation of adaptation goals, models, or rules. Self-improvement which changes the adaptation logic at runtime through meta-adaptation addresses the uncertainty issue.In this paper, we present and discuss a self-improvement case study for an autonomic traffic management system. We adapt parameters of the adaptation logic through rule learning as well as the structure of the adaptation logic, e.g., from central to decentralized control. We show that the resulting implementation enables continuous self-improvement of the system even in situations that have not been taken into account at design time.