Learning relational affordance models for two-arm robots

Abstract

Affordances are used in robotics to model action opportunities of a robotic manipulator on an object in the environment. Previous work has shown how statistical relational learning can be used in a discrete setting to extend affordances to model relations and interactions between multiple objects being manipulated by a robotic arm and deal with environment uncertainty. In this paper, we first extend this concept of relational affordances to a continuous setting and then to a twoarm robot. A relational affordance model can first be learnt for one arm through a behavioural babbling stage, and then with the use of statistical relational learning, after constructing a symmetrical model for the other arm, two-arm manipulation actions can be modelled, where the arms can act sequentially or simultaneously. The model is evaluated in a two-arm action recognition task in a shelf object manipulation setting.

A Survey of Knowledge Representation in Service Robotics

“…The robot's interaction with its environment serves to either learn new motor primitives or skills (trajectory-level) or to learn new properties associated with the type of grasp they make or the skills they use, the object's physical features, and the effects that occur from executing an action (symbolic-level). In works such as [64,65] [66,67,68] [69], a robot can use basic, pre-programmed motor skills (viz. grasping, tapping or touching) to learn about relationships between an object's features (such as shapes, sizes or textures) and features of its actions (such as velocities and point-of-contact).…”

Emergent Structuring of Interdependent Affordance Learning Tasks Using Intrinsic Motivation and Empirical Feature Selection

“…Transfer of learned single-object affordances to bootstrap learning of paired-object affordances was studied by Moldovan et al [18,19]. The authors first used Bayesian Networks (BN) to learn (object, action, effect) relations from single and paired-object interactions, where relative position and orientation were encoded explicitly.…”

“…Different from our model, Moldovan et al focused on generalizing the rules that encode the position and orientation relations between objects; whereas our focus is to learn complex affordances that encode non-linear relations between arbitrary features of objects. In [19], the authors extended their system to continuous settings, however highlevel shape primitives such as cubes and cylinders were predefined, whereas our system can discover such primitives from low-level shape features. Fichtl et al also used predictions of action effects as inputs in predicting effects of other actions [20].…”

Affordance Learning Based on Subtask's Optimal Strategy

“…Reinforcement learning [17,18] Object's manipulation affordance Incremental learning of primitive actions, and context generalization Bayesian network [6,7] Prediction and planning in bi-directional way Statistical relational [14] Model multi-object relationship Ontology knowledge [15,16] Handle object's sudden appear or disappear Support vector machine [9,10,21] Object's manipulation and traversability affordance Prediction and multi-step planning Probability graphical model [19,20] Object's traversability Table 1. Typical learning method under current affordance models…”

“…Hart et al introduced a paradigm for programming adaptive robot control strategies that could be applied in a variety of contexts, furthermore, behavioural affordances are explicitly grounded in the robot's dynamic sensorimotor interactions with its environment [11][12][13]. Moldvan et al employed recent advances in statistical relational learning to learn affordance models for multiple objects that interact with each other, and their approach could be generalized to arbitrary objects [14]. Hidayat et al proposed affordance-based ontology for semantic robots, their model divided the robot's actions into two levels, object selection and manipulation.…”