DataSheet3_Observation vs. interaction in the recognition of human-like movements.zip

A crucial aspect in human-robot collaboration is the robot acceptance by human co-workers. Based on previous experiences of interaction with their fellow beings, humans are able to recognize natural movements of their companions and associate them with the concepts of trust and acceptance. Throughout this process, the judgment is influenced by several percepts, first of all the visual similarity to the companion, which triggers a process of self-identification. When the companion is a robot, the lack of these percepts challenges such a self-identification process, unavoidably lowering the level of acceptance. Hence, while, on the one hand, the robotics industry moves towards manufacturing robots that visually resemble humans, on the other hand, a question is still open on whether the acceptance of robots can be increased by virtue of the movements they exhibit, regardless of their exterior aspect. In order to contribute to answering this question, this paper presents two experimental setups for Turing tests, where an artificial agent performs human-recorded and artificial movements, and a human subject is to judge the human likeness of the movement in two different circumstances: by observing the movement replicated on a screen and by physically interacting with a robot executing the movements. The results reveal that humans are more likely to recognize human movements through interaction than observation, and that, under the interaction condition, artificial movements can be designed to resemble human ones for future robots to be more easily accepted by human co-workers.

人机协作（human-robot collaboration）领域的核心议题之一，是人类协作伙伴对机器人的接受度。基于过往与同类互动的经验，人类能够识别同伴的自然动作，并将其与信任、接纳的认知建立关联。在此过程中，个体的判断受多重感知因素影响，首要因素便是与同伴的视觉相似度——这会触发自我认同的心理过程。当同伴为机器人时，这类感知特征的缺失会阻碍这一自我认同进程，不可避免地降低其被接受的程度。因此，一方面机器人产业正朝着外形更类人的方向研发制造机器人，另一方面仍有一个悬而未决的问题：能否通过机器人的动作表现提升其接受度，而不受其外观形态的限制？为解答这一问题，本文提出了两种图灵测试（Turing test）的实验设置：其中人工智能体（artificial agent）分别执行人类录制的动作与人工生成的动作，人类受试者需要在两种不同场景下判断动作的类人程度：一是观察屏幕上复刻的动作，二是与执行该动作的机器人进行实体交互。研究结果显示，相较于被动观察场景，人类通过实体交互更易识别出人类动作；且在交互场景下，人工动作可被设计得更贴近人类动作模式，从而让未来的机器人更易被人类协作伙伴接纳。