Companies developing autonomous and software-intensive systems show an increasing need to adopt experimentation and data-driven strategies in their development process. With the growing complexity of the systems, companies are increasing their data analytic and experimentation teams to support data-driven development. However, organizations cannot increase in size at the same pace as the system complexity grows. Experimentation teams could run a larger number of experiments by letting the system itself to coordinate its own experiments, instead of the humans. This process is called automated experimentation. However, currently, no tools or frameworks address the challenge of running automated experiments. This paper discusses, through a set of architectural design decisions, the development of an architecture framework that supports automated continuous experiments. The contribution of this paper is twofold. First, it presents, through a set of architectural design decisions, an architecture framework for automated experimentation. Second, it evaluates the architecture framework experimentally in the context of a human-robot interaction proxemics distance problem. This automated experimentation framework aims to deliver more value from the experiments while using fewer R&D resources.