Cooperation and interaction between human and humanoid robots through integration of symbolic expressions and sensorimotor patterns 感覚連動情報とシンボル表現の統合による人とヒューマノイドロボット間の協調と対話に関する研究
Cooperation and interaction between human and humanoid robots through integration of symbolic expressions and sensorimotor patterns
This paper describes a stochastic framework for intelligent humanoidrobots, which can cooperate and interact with humans through integra-tion of symbolic expressions and sensorimotor patterns. The research isdivided into 4 steps. Contributions of the each research step are: 1) anestimation method of sensorimotor patterns of others without having pre-defined user speciffic model in advance through interaction between selfand other, 2) a method to dynamically modify displaying motion pat-terns and to bind the motions with symbol expressions according to per-formance of human-learners, in order for conveying slight differences inmotions, where robotic system coaches humans motions, 3) analysis andmodeling of human-coaches' use of motions and symbolic expressions howthey change them dynamically according to learners performances, and4) demonstration of the feasibility of the robotic motion coaching system,which integrated the methods proposed in step 1) and 2), and the modelsgained in step 3), through experiments of actual sport coaching tasks forbeginners resulted in improvements in motion learning.In the Chapter 1, The main stream of robotics researches are introducedas improvement in individual physical ability. Then, importance of in-telligence of binding symbol expressions and unobservable sensorimotorpatterns, and intelligence to estimate the sensorimotor patterns from ob-servable motions are discussed from interaction point of view.In the Chapter 2, related works are introduced in various fields suchas Robotics, Conversation Analysis, Human-Agent Interaction, Skill andSports Science, and Anticipation of Intention of Others from neuroscienceand cognitive psychology point of view. Then, the chapter addresses chal-lenges from the perspective of required functions for the research. Afterthe discussion of the approach for the resolution method, the Proto-symbolSpace method is introduced as a basic tool for the proposed methods.The Chapter 3 describes an estimation method of sensorimotor patternsof others from motion observation.An approach is to bridge sensorimotor experience, or the Proto-symbolSpaces, between the self and the other. The sensorimotor experience foreach are represented by the Proto-symbol Spaces for each in the research.This approach would result in estimation error due to physical conditiondiffierence between the self and the other. To clear this problem, a methodis proposed in order for adaptive acquisition of Proto-symbol Space ofother by sharing motion patterns and using open questions asking theothers' sensing status described by symbols. Simulation demonstratesthat it is possible to estimate sensorimotor patterns of others with 10-20% errors, even when estimation target motions are not in database.In the second half of the chapter, I discusses about a method to estimateothers' symbol conversion strategy from sensor patterns. The method usesclosed questions asking comparative evaluation of sets of shared motions.The simulation demonstrates that the method can estimate the symbolconversion strategy properly by sharing prepared sets of motions and usingthe closed questions.The Chapter 4 describes a proposing method for dynamic modificationof motion demonstration and for binding the motions with symbol ex-pressions according to performance of human-learners. This method canconvey slight diffierences between learning target motions demonstrated bya coach and motions performed by learners. Feasibility of the method isdemonstrated through experiments of actual sport coaching tasks for be-ginners by using a robotic coaching system. The robotic system coacheshuman-learners tennis forehand swing, by using emphatic motions andadverbial expressions generated from the proposing method. The experi-ments resulted in improvements in motion learning. However, it was notpossible to confirm whether either emphatic motions and/or adverbialexpressions is a contribution factor or not.In the Chapter 5, I discuss about experiments for modeling how human-coaches use emphatic motions and adverbial expressions. In the experi-ments, human-coaches were asked to coach a robot-learner tennis forehandswing, by using the emphatic motions and adverbial expressions. Analysisof the results leads to models; two Adverbial Expression Use Models andtwo Emphatic Motion Use Models.In the Chapter 6, I attempt to integrate the methods proposed in Chapter3 and 4, and the models obtained in Chapter 5. At first, I discuss aboutintegration of the robotic motion coaching system from Chapter 4 andthe models gained from Chapter 5. I then discuss a possible integrationof the method to estimate sensorimotor patterns from the Chapter 3, therobotic motion coaching system from Chapter 4, and the models gainedfrom Chapter 5.I demonstrated the feasibility of the robotic motion coaching system inte-grated with one of the EMU-Model and one of the AEU-Model, by experi-ments of a tennis forehand swing coaching task for beginners. I confirmedthat the EMU-Model and the AEU-Model contribute to improvement inmotion learning. It is demonstrated that value output by the EMU-Modelis a contribution factor by a statistic analysis. I also found there is animprovement in motion learning when using the AEU-Models. However,even though I found positive contribution of the adverbial expressions forthe improvement in motion learning, it is not able to decide whether theadverbial expressions chosen by using the AEU-Model is a contributionfactor or not.The thesis is then concluded in the Chapter 7.