Improving the adaptability of simulated evolutionary swarm robots in dynamically changing environments

Area of Science:

• Robotics
• Evolutionary Computation
• Bio-inspired Systems

Background:

• Adapting to fluctuating environments is a key challenge in evolutionary robotics.
• Current systems often require complete re-optimization for new conditions.

Purpose of the Study:

• To explore the adaptive potential of simulated swarm robots using genomic encoding of bio-inspired gene regulatory networks (GRNs).
• To investigate how separating static and conditionally active network components impacts adaptive behavior.

Main Methods:

• Utilized an artificial genome combined with a flexible agent-based system in a dynamic artificial life simulation.
• Implemented GRNs that transduce environmental cues into phenotypic behavior.

Main Results:

• Separating static and conditionally active network parts improved adaptive behavior.
• The system stores previously optimized GRNs, enabling faster re-adaptation to re-encountered environmental conditions.
• This contrasts with ANN-based systems needing complete re-optimization.

Conclusions:

• Evolutionary principles applied to GRNs accelerate and enhance adaptation in non-stable environments.
• Storing and selectively activating GRNs facilitates efficient re-adaptation.
• This bio-inspired genomic approach offers a promising alternative for adaptive robotics.