Challenges in developing software for cyber-physical systems

Cyber-physical systems are systems that integrate the digital computational world with the real physical world, often using sensors and actuators as interfaces. There exist many application domains of cyber-physical systems such as autonomous systems, process control systems, robotic systems, and context-aware systems. The physical world is a complex and continuous world that changes in real-time while the computational world is a simplified and discrete world that often stores a delayed, likely inaccurate image of the physical world using sensory data. The mismatch between these two worlds poses unique challenges of developing software for cyber-physical systems. In this paper, we mainly (1) illustrate this mismatch problem by presenting difficulties that we faced during two projects of developing software for cyber-physical systems: the self-adaptive robot car project and iSpace project; (2) systematically analyse the reason for the mismatch problem based on our first-hand experiences; (3) summarize a general methodology on how to partly bridge the gap between the two mismatched worlds to better develop software for cyber-physical systems based on the knowledge learned from the two projects. We expect that these experiences can help others in better design and develop cyber-physical systems and can be used to explore new techniques (such as new testing techniques) to improve the quality of cyber-physical systems.