Spatial reasoning for intelligent action and for human-robot interaction

摘要

Successful and robust communication between cognitive agents, inparticular between humans and robots, requires shared concepts betweenthe communicating agents. In my lecture, I will present spatial robotenvironments as an example for domains of interaction between cogni-tive agents in which spatial information to be communicated needs to bephysically grounded in the environment. I will give examples of spatialcommunication tasks and discuss levels of abstraction for spatial infor-mation and knowledge from the physical level. I will show how spatialcontext can be exploited for cognitive off-loading to make communica-tion effective and efficient. With this approach, I will motivate the situ-ation-dependent selection of appropriate levels of conceptualization forspatial knowledge. The talk will present qualitative spatial abstraction asa suitable approach to establish a conceptual level of spatial knowledgethat supports an important class of communication and reasoning tasks. Iwill discuss how this conceptual level relates to the more commonquantitative representations of spatial robot environments for autono-mous robots. I will propose ways in which the different forms of repre-sentation may be combined for fruitful interaction. I will discussadvantages and disadvantages of various forms of representation fordifferent tasks and purposes. In the talk, I will present examples of spatialreasoning with qualitative representations and illustrate importantproperties of these representations. Navigation tasks will serve asexample for tasks in which detailed quantitative planning may beundesirable. Although in some sense, qualitative representations aremore abstract than quantitative representations, in another sense they aremore concrete, as they maintain important structural properties of spatio-temporal events that are relevant to spatial cognition. This raises the issueof 'cognitive adequacy' of spatial representations, which I will alsoaddress in my talk. It also raises the issue of the respective expressivepower of qualitative versus quantitative analysis. Finally, my talk willaddress the relation between spatial and linguistic forms of representingspatial knowledge, for example, maps and textual descriptions of space:they can be viewed as alternative or complementary ways of communi-cating knowledge, each having their specific advantages.