Affordance-Based Human-Robot Interaction
In our targeted scenario, humans can flexibly establish joint object reference with a robot entirely on the basis of their own intuitions. To reach this aim, the robot needs to be equipped with the kind of knowledge that can be matched in a cognitively ad
- PDF / 2,918,809 Bytes
- 14 Pages / 430 x 660 pts Page_size
- 95 Downloads / 217 Views
stract. In our targeted scenario, humans can flexibly establish joint object reference with a robot entirely on the basis of their own intuitions. To reach this aim, the robot needs to be equipped with the kind of knowledge that can be matched in a cognitively adequate way to users’ intuitive conceptual and linguistic preferences. Such an endeavour requires knowledge about human spatial object reference under consideration of object affordances and functional features. In this paper we motivate our approach by reviewing relevant insights gained in the field of Spatial Cognition, and we discuss the suitability of our robotic system to incorporate these findings. In our context, affordances are visually perceivable functional object aspects shared by the designer of the recognition module and the prospective robot user or instructor.
1 Introduction Human-robot interaction centers on events in which human instructors expect robots to perform desired actions on specified objects [Zhang and Knoll, 2003]. A shared goal across contexts is therefore to establish joint reference to one or several objects present [Moratz et al., 2001]. This goal can be achieved successfully in one of several ways: for example, instructors could specify exact metric measures, knowledge about all potentially needed objects could be implemented or successively taught to the robot, or users could be provided with a list of object names or class IDs that the robot can understand. However, each of these methods comes with its own problems. In more complex settings or open scenarios, and whenever generic tasks need to be formulated, the limits of predefined referring strategies become obvious. Generic tasks need to be specified by a set of rules comprising complex robot commands. An everyday example involving a future service robot is the following. The robot could be taught by an untrained user to set the table as follows: Each cutlery piece (knives, forks etc.) is placed at the side of the plate which corresponds to the side where the human hand is that will use the tool. If pasta is served, the fork would be on the right side of the plate, otherwise it would typically be on the left side. If these principles are taught to the robot as a set of functional rules, the robot can generalize to a new scenario in a sensible manner. For example, if a guest for some reason can use only one arm, all cutlery needs to be placed at the corresponding side. Also, culturally diverse habits can easily be accounted for. Such generic rules can be formulated by linguistic functional propositions as instruction representations. The problem addressed in this paper is how to let the service robot acquire coarse, underspecified knowledge from the environment, which is functionally motivated E. Rome et al. (Eds.): Affordance-Based Robot Control, LNAI 4760, pp. 63–76, 2008. c Springer-Verlag Berlin Heidelberg 2008
64
R. Moratz and T. Tenbrink
[Hois et al., 2006] and matches natural human strategies of spatial reference. Since we consider it essential to avoid the us
Data Loading...
