Modular robotics link the reliability of a centralised system with the adaptivity of a decentralised system. It is difficult for a robot with a fixed shape to be able to perform many different types of tasks. As the task space grows, the number of functions integrated into a robot scales, which increases the number of compromises in design. By separating the robot into individual autonomous agents, a distributed system is able to adapt to changes in the environment or task. Modular robotics create highly adaptable robots through using a set of interlocking components to form a complete system. However, reliability becomes fundamental when scaling in number. As each module is self-contained with on-board sensing, power, actuation, and communication, if one of these elements were to fail, the module's capabilities are compromised. A module failing disables some, if not all, of the robot's functionality. Increasing resilience is essential for the applicability of modular robotics, as their failures have the frequency of a distributed system but the severity of a centralised system. Regardless of how robust an individual module is, increasing the number of individual agents in any robotic system reduces the odds of completing a mission without failure. Robustness must encapsulate not just the module itself, but rather extend to the interactions between modules. Improving the reliability of an adaptive system builds sustainability, as a single platform can perform many tasks, mitigating waste and downtime. This thesis investigates fundamental topics on robustness of modular robotic systems: platforms, control, and sharing. In order to increase the number of modules, the base platform first needs to be reliable. Modules are independent agents and need considerations for autonomy, but also construct the overall system and have requirements according to the robot's functionality. Within this context, I introduce a new modular robot platform resembling polygon meshes, Mori3. Mori3 acts as a bridge between modular and origami robots, applying the benefits of origami robots to the self-reconfigurable modular robot~(SRMR) domain. I demonstrate Mori3 in diverse scenarios, using quadrupedal locomotion, closed chain rolling, object manipulation, and variable surfaces applications. Control of modular robots is addressed, spanning from low-level to high-level interactive control for arbitrary topologies. Robust modular robotics consist of determined behaviours which span the whole robot, but also contain emergent behaviours which come from the relationships between modules. I propose strategies for sharing communication protocols across SRMRs to improve the consistency and control. I introduce a modular control platform, consisting of physical reconfigurable joysticks and an optimization scheme to prevent dangerous robot actions. Merging trajectory optimisation and reconfigurable joysticks allows different modular robot structures to perform new tasks without a prior task-based controller. Lastly, I introduce sharing as a foundational topic in modular robotics. When scaling in number, adding modules increases the number of components which can fail. However, adding modules does not just increase the odds that a singular component fails, but also increases the number of usable components across the overall robot. Sharing takes a module's resources, such as data and power, and integrates them across the span of the system through l