Selective Object Aggregation using Swarm Robotics

Resumo

Selective object aggregation refers to the autonomous clustering of objects based on specific attributes. In this context, swarm robotics, inspired by the collective behavior of natural systems, seeks to control a group of simple agents to locate and organize randomly distributed objects by similarity, without centralized control or explicit inter-agent communication. In this work, we propose an adaptation of the Lumer & Faieta algorithm to control a swarm of robots for the selective aggregation of objects by type. The approach is inspired by the self-organizing behavior observed in ant colonies, which naturally sort debris and dead bodies into clusters to maintain nest hygiene. The algorithm is chosen for its efficiency in grouping items based solely on local information, enabling decentralized operation. We present an enhanced version of this algorithm, tailored for robots with limited perception and decision-making capabilities. Despite these individual limitations, the swarm collectively achieves the formation of well-organized clusters. The implementation relies on local neighborhood perception, where the decision to pick up or drop an object is governed by the density of similar items in the robot’s immediate vicinity. Adapted probability functions, derived from the original Lumer & Faieta formulation, encourage isolated objects to be picked up and deposited in regions with a higher concentration of similar types, gradually fostering cluster formation. The system is implemented in the CoppeliaSim simulation environment, utilizing Khepera-III mobile robots, chosen for their compact design and versatile sensor suite. Each robot is equipped with a gripper for object manipulation and a vision sensor for object recognition, allowing it to identify color and classify object type. To evaluate the robustness of the proposed method, we test various arena discretizations and swarm sizes. Key performance metrics include the execution time during the search and aggregation phases, as well as the quality and coherence of the final clusters. Simulation results confirm the effectiveness of the approach, demonstrating the system’s ability to reproduce clustering patterns akin to those observed in biological systems. The swarm maintains selective aggregation throughout the process, underscoring the potential of simple local interactions to generate sophisticated collective behaviors.