Planning sensing strategies in a robot work cell with multi-sensor capabilities

Abstract

In this paper we present an approach to planning sensing strategies in a robot work cell with multi-sensor capabilities. The system first forms an initial set of object hypotheses by using one of the sensors. Subsequently, the system reasons over different possibilities for selecting the next sensing operation, this being done in a manner so as to maximally disambiguate the initial set of hypotheses. The "next sensing operation" is characterized by both the choice of the sensor and the viewpoint to be used. Aspect graph representation of objects plays a central role in the selection of the viewpoint, these representations being derived automatically by a solid modelling program.

Active recognition through next view planning: a survey

“…Dickinson et al [56] use a variant of a Bayes net for their recognition system (Rimey and Brown [25]. use Bayes nets for scene analysis), while Hutchinson and Kak [55] use the Dempster-Shafer theory to represent uncertainty. Some scene analysis systems use fuzzy logic (e.g., [69]).…”

“…Many active object recongition schemes are based on aspect graphs [55], [56], [14], [57], [58] (Section 2.3 describes these in detail). The aspect graph-based approach is more general than the other two approaches in that appearance-based information may be used to construct an aspect graph.…”

“…In their work on planning sensing strategies in a robot work cell with multi-sensor capabilities, Hutchinson and Kak [55] use an aspect graph to represent information about the objects in their model base. They present a system for dynamically planning sensing strategies, based on the current best estimate of the world.…”

“…While many approaches such as those of Hutchinson and Kak [55] and Liu and Tsai [70] use geometric features, the scheme of Gremban and Ikeuchi [14] is independent of the features used. The latter present results with geometric and photometric information.…”

“…Dickinson et al [56] use a hierarchical representation scheme based on volumetric primitives. Due to the non-hierarchical nature of Hutchinson and Kak's system [55], many redundant hypotheses are proposed, which have to be later removed through consistency checks. In our work on aspect graph-based modeling and recognition, the hierarchical knowledge representation scheme and probabilistic hypothesis generation mechanism itself refines feature evidence through different lev-els -leading to simpler evidence propagation and less computational cost [57], [58].…”