RESUMO
Robots artificially replicate human capabilities thanks to their software, the main embodiment of intelligence. However, engineering robotics software has become increasingly challenging. Developers need expertise from different disciplines as well as they are faced with heterogeneous hardware and uncertain operating environments. To this end, the software needs to be variable-to customize robots for different customers, hardware, and operating environments. However, variability adds substantial complexity and needs to be managed-yet, ad hoc practices prevail in the robotics domain, challenging effective software reuse, maintenance, and evolution. To improve the situation, we need to enhance our empirical understanding of variability in robotics. We present a multiple-case study on software variability in the vibrant and challenging domain of service robotics. We investigated drivers, practices, methods, and challenges of variability from industrial companies building service robots. We analyzed the state-of-the-practice and the state-of-the-art-the former via an experience report and eleven interviews with two service robotics companies; the latter via a systematic literature review. We triangulated from these sources, reporting observations with actionable recommendations for researchers, tool providers, and practitioners. We formulated hypotheses trying to explain our observations, and also compared the state-of-the-art from the literature with the-state-of-the-practice we observed in our cases. We learned that the level of abstraction in robotics software needs to be raised for simplifying variability management and software integration, while keeping a sufficient level of customization to boost efficiency and effectiveness in their robots' operation. Planning and realizing variability for specific requirements and implementing robust abstractions permit robotic applications to operate robustly in dynamic environments, which are often only partially known and controllable. With this aim, our companies use a number of mechanisms, some of them based on formalisms used to specify robotic behavior, such as finite-state machines and behavior trees. To foster software reuse, the service robotics domain will greatly benefit from having software components-completely decoupled from hardware-with harmonized and standardized interfaces, and organized in an ecosystem shared among various companies.
RESUMO
Context: During the first wave of the COVID-19 pandemic, an international and heterogeneous team of scientists collaborated on a social project to produce a mechanical ventilator for intensive care units (MVM). MVM has been conceived to be produced and used also in poor countries: it is open-source, no patents, cheap, and can be produced with materials that are easy to retrieve. Objective: The objective of this work is to extract from the experience of the MVM development and software certification a set of lessons learned and then guidelines that can help developers to produce safety-critical devices in similar emergency situations. Method: We conducted a case study. We had full access to source code, comments on code, change requests, test reports, every deliverable (60 in total) produced for the software certification (safety concepts, requirements specifications, architecture and design, testing activities, etc.), notes, whiteboard sketches, emails, etc. We validated both lessons learned and guidelines with experts. Findings: We contribute a set of validated lessons learned and a set of validated guidelines, together with a discussion of benefits and risks of each guideline. Conclusion: In this work we share our experience in certifying software for healthcare devices produced under emergency, i.e. with strict and pressing time constraints and with the difficulty of establishing a heterogeneous development team made of volunteers. We believe that the guidelines will help engineers during the development of critical software under emergency.
