Adaptive Behavior of Collaborative Robots: Review and Investigation of Human Predictive Ability

With the increasing demand for product variety and customization in manufacturing, the need for flexible production and assembly systems, particularly for small and medium-sized enterprises (SMEs), is growing. Collaborative human-robot assembly (HRA) has emerged as a promising solution to enhance competitiveness by leveraging the strengths of both humans and robots. However, the inherent variability in human performance can disrupt the smooth interaction and processes in HRA systems. Artificial intelligence (AI) methods offer opportunities to develop adaptive robots that can dynamically adjust to the behavior and needs of their human partners. While the adaptivity of collaborative robots promises significant advantages, it may also introduce challenges in predicting the robot's behavior and actions from a human perspective. Uncertainty due to ambiguity and confusion arising from the difficulty in understanding the automatisms of the robot's actions could lead to psychological stress for human partners, impacting their well-being, trust, acceptance, and performance. Especially as production tends to integrate the concept of Industry 5.0, which focuses on the development of human-centered systems, addressing these challenges becomes crucial. This paper provides a comprehensive review of previous work on the adaptive behavior of collaborative robots, aiming to define the concept and identify the variables and mechanisms that make robot behavior adaptive. The review is completed by a study with 40 participants to investigate the effect of the degree of adaptation in robot behavior on human predictive ability. The results showed a significant effect, with an average matching of only 49% between participants' expectations and the robot's actual adaptation actions. Participants with previous robot experience showed improved but limited predictive ability (20% relative increase). The findings highlight significant difficulties in anticipating adaptive robot behavior, especially in more complex adaptive scenarios and for decision-making processes, underscoring the need for enhanced transparency to facilitate better human understanding and acceptance in collaborative robotic manufacturing environments.