Adaptive Design in Human-Robot Collaboration

Abstract

Artificial intelligence (AI) and sensor technology have made collaborative robots (cobots) more relevant in the industry. However, an understanding of how human-cobot interaction differs from human-human interaction is vital, as cobots become more adaptive to workers' behaviour and needs. We conducted a controlled and incentivized "field-in-the-lab" experiment in a realistic production environment. In all treatments, there was a human Worker 1 at Station 1 in a two-station production line. We varied whether Worker 2 at Station 2 was a human or a robot. The teams produced electronic motor components, and Worker 1 could submit intermediate products to Worker 2 or a "waiting queue." All components that completed the queue or were finished by Worker 2 counted toward the team payoff. Worker 1's production speed was classified as "high" (H), "medium" (M), or "low" (L). In line with this speed, in the “adaptive” treatments, the robot was set to production speed H>M>L, Worker 2 received productivity feedback H>M>L, or Worker 2 received piece rate H>M>L. We find strong robotic aversion and blame shifting toward the robots. Adaptivity increased productivity and mildly reduced robotic aversion. It also reduced the allocation of blame, but this reduction was not statistically significant. This study has important implications for using cobots and can inform the adaptive design of effective human-robot collaboration, which can enhance workplace productivity.