Artificial Life Laboratory, Graz, Austria

  The new field of self-reconfiguring modular robotics (i.e.,
  decentrally controlled `super-robots' based on autonomous,
  interacting robot modules with variable morphologies) calls for new
  paradigms of designing robot controllers. One option is the approach
  of evolutionary robotics. The challenge is to achieve high
  evaluation numbers with the available resources which may even
  affect the feasibility of the approach. Simulations are applied at
  least in a preliminary stage of research to lower these
  costs. However, even simulations are computationally expensive which
  gets even more burdensome once comprehensive studies and comparisons
  between different controller designs and approaches have to be
  done. Hence, a benchmark with low computational cost is needed that
  still contains the typical challenges of decentral control, is
  comparable, and easily manageable. We propose such a benchmark and
  report an empirical study of its characteristics including the
  transition from the single-robot setting to the multi-robot setting,
  typical local optima, and properties of adaptive walks through the
  fitness landscape.