Influence of an Altruism Parameter in Active Inference Agents on One-to-Two Passing Behaviour

The societal deployment of autonomous mobile robots necessitates navigation designs that can handle interactions with multiple people. This study extends a prior altruism model based on active inference to multi-agent scenarios and investigates the influence of altruism parameters on passing behavior. We constructed a simulation model of a one-on-two passing situation and evaluated three agent models: a baseline model that considers all agents, a model that selectively considers agents within a certain physical distance, and a model that adjusts its altruism level based on distance. The simulation results indicate that limiting the scope of altruistic consideration to agents within a defined physical distance can reproduce behavior equivalent to that of the comprehensive model which considers all agents. This finding suggests a crucial design principle for mobile robots that balances the benefits of altruistic behavior with the reduction of computational resources, a significant challenge as the number of interacting agents increases.