Emergent Structures in Modular Self-Reconfigurable Robots

We demonstrate how simple local sensing and control rules achieve useful emergent behaviors in modular self-reconfigurable (metamorphic) robots. Our biologically inspired approach grows structures with the desired functionality even though the final shapes have some unspecified, random variation. By contrast, other self-reconfiguration algorithms require an a-priori exact description of a target shape for the given task, which may be difficult when a robot operates in uncertain environments. We present and evaluate several control algorithms through simulation experiments of Proteo, a metamorphic robot system. 1 I n t r o d u c t i o n Modular self-reconfigurable (or metamorphic) robotics is a relatively new concept [3, 9, 4, 8, 7, 5, 6]. Such robotic systems consist of many simple identical modules, that can attach and detach from one another to change their overall topology. These systems can dynamically adapt their shape to suit the needs of the task at hand, e.g., for manipulation, locomotion, and creation of static structures. The fact that all modules are identical results in robustness with respect to a task: because no module is critical, if any break down they can be easily replaced by others. In addition, because typical systems consist of a large number of modules (hundreds or thousands) reductions in manufacturing costs can be brought about through mass production. Prom a planning and control viewpoint, metamorphic robots pose several interesting research challenges. Self-reconfiguration, or how to change shape automatically, is a new and so far little studied problem. Decentralized control is a usefifl approach to this problem due to the large number of modules in a typical robot, and the fact that each module is a selfcontained unit with its own processing, sensing and "Contac t information: acasa l~parc .xerox .com 0 -7803 -5886 -4 /00 /$ 10 .00© 2000 IEEE 1 cha Casal~ Tad Hogg enter , Pa lo Alto , CA 94304 actuation. This observation means the large body of work on distributed multi-agent control is particularly relevant. The problem of self-reconfiguration is unique to metamorphic robots 1. Ideally, reconfiguration should occur in the minimum number of steps (module moves). Such an optimal solution to reconfiguration, a combinatorial optimization problem, involves searching the space of all possible reconfiguration sequences for the optimal one. The space of robot configurations corresponds to all the possible arrangements of a set of labeled connected modules, which grows exponentially with the number of modules [15]. Therefore, for robots with hundreds of modules or more, the size of the search space makes finding a globally optimal solution intractable. A number of algorithms use heuristics to find nearoptimal solutions for different robot systems [9, 4, 6, 7, 8, 10, ll]. These methods all require determining and defining, ahead of time, a desired shape appropriate to solve the task at hand. This target shape becomes the goal input to the reconfiguration algorithm and requires an exact (geometrical or otherwise) description. However, there are instances when defining an exact target shape may not be suitable or even possible. This may arise when there is uncertainty about the environment or the task, for example, when grasping an object of unknown size or shape. Our approach differs from these self-reconfiguration algorithms in several respects. Even though reconfiguration is achieved, the algorithm does not aim at attaining an exact predefined shape. Rather, the goal is to create a structure with the correct properties (structural, morphological, etc) required to achieve the task. Any stable "emergent" structure that exhibits the desired properties is considered satisfactory, with no regard for the "optimality" or details of the resulting geometry. In this respect, our approach is more akin to the biologically-inspired ideas found in the Artificial Life field. Distributed or multi-agent control is applied to 1 a similar problem is the classic Transpor ta t ion problem [20].