Agent Sensing In Limited Resource Environments

One of the key challenges for multiagent systems (MAS) is optimizing performance in limited resource environments. Previous research in this area has focused on the problems of 1) resource allocation and arbitration, and 2) bounded rationality, which describe the relationship between resource constraints and both agent reasoning and actuation. However, less work exists addressing the effect of consuming resources during agent sensing, particularly two important tradeoffs. First, sensing can reduce resource availability, resulting in a tradeoff between overall system performance and an agent’s sensing behavior (the Performance Tradeoff). Second, consuming resources during sensing can alter the outcome of the measurement (the Observer Effect). Since an agent requires up-to-date information, but tracking too frequently can worsen the observer effect, there also exists a tradeoff between the quality and frequency of an agent’s sensing (the Information Quality Tradeoff). We present an algorithm for Resource-Aware Tradeoff-based Sensing (RATS) which considers trends in both the need for information and system performance to learn an appropriate sensing frequency. The agent considers a sliding window of possible frequencies bounded to avoid decreases in system performance while providing quality information and chooses an appropriate frequency based on its confidence in sensing. To validate our algorithm, we conducted experiments with 30 agents in a simulation of agent-based wireless networks, with different levels of resource constraints, to compare RATS sensing against only-need-aware sensing and only-performance-aware sensing. Our results show that RATS agents experience better system performance than only-need-aware sensing, while producing more accurate models than only-performance-aware sensing.