Designing robotic systems that can autonomously interact with their environment to solve tasks remains a major challenge. Conventional approaches use multi-sensor feedback loops to control displacements. This is often coupled with algorithmic and hardware complexity, which tend to be detrimental to energy efficiency, reliability, and form factor, all key aspects in autonomous systems. Here we I will present a new decentralized and modular robotic platform that I have developed, Granulobot, to demonstrate tasks based on aggregate mechanical properties. Granulobots consist of active, gear-like particles that magnetically interact with each other and produce torques. They can self-assemble into aggregates that can reconfigure in real-time. The apparent complexity of a system with many degrees of freedom is made an advantage by leveraging the material-like properties of aggregates. In particular, aggregates can transition between rigid and liquid-like states with a wide range of effective viscosities. This enables the robot to move in complex environments through holes and over obstacles by setting a single “material” parameter, and without centralized control or real-time sensor-based feedback. Such minimal control, enabled by the modular design of Granulobots, advances robotic autonomy by exploring a morphological form of computation and feedback that greatly reduces the number of control parameters that must be taken care of by the embedded systems or operators.