Validation of whole-body loco-manipulation affordances for pushability and liftability

Abstract

Autonomous robots that are intended to work in disaster scenarios like collapsed or contaminated buildings need to be able to efficiently identify action possibilities in unknown environments. This includes the detection of environmental elements that allow interaction, such as doors or debris, as well as the utilization of fixed environmental structures for stable whole-body loco-manipulation. Affordances that refer to whole-body actions are especially valuable for humanoid robots as the necessity of stabilization is an integral part of their control strategies. Based on our previous work we propose to apply the concept of affordances to actions of stable whole-body loco-manipulation, in particular to pushing and lifting of large objects. We extend our perceptual pipeline in order to build large-scale representations of the robot's environment in terms of environmental primitives like planes, cylinders and spheres. A rule-based system is employed to derive whole-body affordance hypotheses from these primitives, which are then subject to validation by the robot. An experimental evaluation demonstrates our progress in detection, validation and utilization of whole-body affordances.