Evidence accumulation with adaptive weighting of social and personal information for collective perception

Abstract This study advances collective perception in swarm robotics by introducing the Automated Swarm Opinion Diffusion Model (aSODM), which addresses the limitations of the original Swarm Opinion Diffusion Model (SODM). Both aSODM and SODM are evidence accumulation methods based on human-like decision-making processes. While SODM integrates social and personal information similarly to aSODM, its reliance on a manually predetermined social factor parameter limits its adaptability to diverse task difficulties. The Automated Swarm Opinion Diffusion Model eliminates this dependency by introducing an adaptive personal factor parameter, which automatically adjusts the weighting of personal and social information based on the information gathered about the environment. This automated approach improves robustness and reduces the dissemination of erroneous information in the early phases of a task. Comparative simulations against baseline methods (Voter Model and Majority Rule) demonstrate that aSODM enhances efficiency, particularly in tasks with higher difficulty levels. While aSODM outperforms SODM, its automated critical parameter selection makes it particularly well-suited for real-world applications where prior knowledge of the environment and task difficulty is lacking.