Detecting Contingency for HRI in Open-World Environments

摘要: This paper presents a novel algorithm for detecting contingent reactions to robot behavior in noisy real-world environments with naive users. Prior work has established that one way detect contingency is by calculating difference metric between sensor data before and after probe of the environment. Our algorithm, CIRCLE (Contingency Interactive Real-time CLassification Engagement) provides new approach this contingency, improving running time calculation from 2.5 seconds approximately 0.001 on an 1100-sample vector, effectively enabling real-time detection events. We show accuracy comparable best offline results (89.5% vs 91% prior work), demonstrate utility field study survey-administering open-world environment users, showing can decrease number requests it makes (from 38 13) while more efficiently collecting survey responses (30% response rate rather than 26.3%).