Research Of Non-cooperative Human-robot Interaction Based On Humanoid Robot

Human-robot interaction (HRI) is an extensively studied topic in machine learn-ing,artifcialintelligenceandrobotics. AsHRIbecomesmoreandmoremature,peopleare wondering whether robots could replace humans in complex and hazardous jobs orplay a role in human society. There are two kinds of HRI: cooperative interaction andnon-cooperative interaction. Cooperative interaction is to improve the efciency for arobottofulfllprescribedtasksviapromptevaluationsonitsperformance. Manyefortsare focusing on cooperative interaction, from individual use to military felds. Most in-teractions between humans in real life are non-cooperative. Incited by the motivationthat intelligent robots imitate human-human interaction, research on non-cooperativehuman-robot interaction (NCHRI) is considered to be necessary and urgent. However,NCHRI has not received enough attention. If humans have a negative impact on taskcompletion, whether robot could complete tasks efectively. Or if humans play a rolein preventing robots from accomplishing goals in HRI, whether robot could recognizehuman’s intention intelligently to overcome users’ impediment is a critical issue to beinvestigated in robotics.Here for one kind NCHRI, in which humans play a role in preventing robots fromaccomplishing goals, a strategy for modeling human’s intention online is proposed.Then robots can be much smarter in case of human’s resistance and then recognizehuman’s intention. A mora game is adopted to verify the availability of the method,which is based on the improved Bayesian inference. A fast hand gesture recognitionmethod based on geometric features is proposed to fulfl the requirement of naturalinteraction. Experiments on NAO humanoid robot platform show the validity of theproposed method. The major work of this thesis is organized as follows:1. Proposed a strategy for modeling human’s intention online for Inference of user’sintention for non-cooperative Human-robot interaction. A mora game is adopted toverify the availability and performance of the method. Designed the frst humanoidrobot which could beat huamsns in more games.2. A fast hand gesture recognition method based on geometric features is proposed tofulfl the real-time and accuracy request of interaction.3. Designed a complete process of human-computer interaction. The whole processof interaction depends on natural interactive channels (voice, visual, tactile). Accom-plished the whole NCHRI system.