The emerging ubiquitous computing services integrate numerous distributed and heterogeneous components, which incur significantly high costs for their development, maintenance and administration. In this paper we introduce a middleware architecture for ubiquitous contextaware services which eases integration, while also including a wide range of features that maximize service autonomy. Autonomy is addressed at various levels, including context-acquisition components, situation modeling components and services. Several of these components are implemented as software agents given the advantages of agent technologies for realizing service autonomy. Along with these agents, adaptive perceptive interfaces ensuring autonomy at the context-acquisition level have been developed and integrated with the agent societies. The introduced architecture deals primarily with self-healing (recovery) and self-configuration (adaptation) characteristics of the typical autonomic systems. Following the illustration of the framework, we elaborate on how it has been used to support realistic prototype context-aware human centric and non-obtrusive services.