A number of initiatives promote the topic of digital manufacturing, i.e., the adoption of pervasive, fine-grained cyber-physical systems (CPS) and of data-driven optimization techniques within the manufacturing domain, both in order to gain efficiency while running conventional operations and to allow for new production and business models such as mass customization or networked value chains.In this context, we contribute the findings from our experience in pursuing this vision through two projects with different goals (gradual improvement over current operations vs. more ambitious rethink), target maturity of their results (production-grade vs. proof-of-concept pilots), and therefore approaches. Concretely, we first elaborate on the often neglected line-level automation and the importance of making it a first-class citizen in digital manufacturing, especially for transitioning current manufacturing facilities into future scenarios. We describe the approach and reference the specific technologies used for its production deployment on the factory floor at a customer organization. We then introduce the positioning, within a second stack jointly prototyped with partner organizations, of reasoning algorithms. As entry points for added autonomy into manufacturing systems, they pave the way for automatic (re)configuration from high level goals and for error recovery.This contribution provides insights on how to design the next generation of software architectures for manufacturing, and aims to contribute to the transition from early reference architectural models into design blueprints for actual technology stacks.

We consider the problem of cooperative localization (CL) via interrobot measurements for a team of networked robots with limited on-board resources. We propose a novel algorithm in which each robot localizes itself in a global coordinate frame by local dead reckoning, and opportunistically corrects its pose estimate whenever it receives a relative measurement update message from a server. The computation and storage cost per robot in terms of the size of the team is of order $O(1)$, and the robots are only transmitting information when they are involved in a relative measurement. The server also only needs to compute and transmit update messages when it receives an interrobot measurement. Under perfect communication, our algorithm is an alternative implementation of a joint CL for the team via an extended Kalman filter. However, perfect communication is not a hard constraint. We show that our algorithm is robust to communication failures, with formal guarantees that the updated estimates of the robots receiving the update message are of minimum variance in a first-order approximate sense at that given timestep. We demonstrate the performance of our algorithm in simulation and experiments.