Reconfiguration Planning of Self-Robotic System

The transforming task of a modular system from one configuration into another is called the Reconfiguration Planning problem. Solving this problem is fundamental to any SR system. In some approaches explicit start and goal configurations are given, and in others the goal shape is defined by desired properties. Centralized algorithms require global system knowledge and compute reconfiguration plans directly, whereas decentralized algorithms compute solutions in a distributed fashion without the use of a central controller.

Reconfiguration algorithms can be designed for classes of modules, or for specific robots. Often a centralized solution is more obvious and is developed first, followed by a distributed version, although not always. Not all decentralized algorithms are guaranteed to converge to a solution, or are correct for arbitrary goal shapes.

Reconfiguration of CEBOT was planned by a central control cell known as a master. Master cells were later intended to be dynamically chosen, blurring the distinction between centralization and decentralization. Later CEBOT control is hierarchical (behavior-based). A common technique is used in reconfiguration algorithms for a lattice-based system is to build a graph representation of the robot configuration, and then to use standard graph techniques such as search to compute motion plans. Planning for the Molecule robot developed by the Dartmouth group is one example. Other example from the Dartmouth group is planning for unit-compressible systems such as the Crystal. This planner, named MeltGrow, uses the concept of a metamodule , where a group of modules are treated as a single unit with additional motion capabilities. The Crystal robot implements convex transitions using metamodules called Grains. Graph-based algorithms are also used by the MTRAN planner to compute individual module trajectories.

Pre-computed data structures can also be centralized by planners store such as gait-control tables. Once a gait is selected by the central controller, it is executed by local controllers on the individual modules. This type of algorithm is used by Polypod. The division between central and local controllers is also used in by RMMS, and I-Cubes.