Pervasive computing environments such as smart spaces require a mechanism to easily integrate, manage and use numerous, heterogeneous sensors and actuators into the system. However, available sensor network platforms are inadequate for this task. The goals are requirements for a smart space are very different from the typical sensor network application. Specifically, we found that the manual integration of devices must be replaced by a scalable, plug-and-play mechanism. The space should be assembled programmatically by software developers, not hardwired by engineers and system integrators. This allows for cost-effective development, enables extensibility, and simplifies change management. We found that in a smart space, computation and power are readily available and connectivity is stable and rarely ad-hoc. Our deployment of a smart house (an assistive environment for seniors) guided us to designing Atlas, a new, commercially available service-oriented sensor and actuator platform that enables self-integrative, programmable pervasive spaces. We present the design and implementation of the Atlas hardware and middleware components, its salient characteristics, and several case studies of projects using Atlas. © 2006 IEEE.