Interactive object classification using sensorimotor contingencies

Högman, Virgile; Björkman, Mårten; Kragić, Danica

doi:10.1109/iros.2013.6696752

Cited by 15 publications

(19 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…More recently, the area of interactive perception goes beyond the mere selection of percepts toward actively changing the state of the environment to increase information gain. In studies on robot vision, it has been shown that a visual scene can be disambiguated by actively manipulating objects in the scene through a robotic arm (Fitzpatrick and Metta 2002;Björkman et al 2013;Högman et al 2013). Similar implications apply to other sensory modalities.…”

Section: Action In Robotic Perceptionmentioning

confidence: 92%

Implications of Action-Oriented Paradigm Shifts in Cognitive Science

et al. 2016

View full text Add to dashboard Cite

An action-oriented perspective changes the role of an individual from a passive observer to an actively engaged agent interacting in a closed loop with the world as well as with others. Cognition exists to serve action within a landscape that contains both. This chapter surveys this landscape and addresses the status of the pragmatic turn. Its potential infl uence on science and the study of cognition are considered (including perception, social cognition, social interaction, sensorimotor entrainment, and language acquisition) and its impact on how neuroscience is studied is also investigated (with the notion that brains do not passively build models, but instead support the guidance of action).A review of its implications in robotics and engineering includes a discussion of the application of enactive control principles to couple action and perception in robotics as well as the conceptualization of system design in a more holistic, less modular manner. Practical applications that can impact the human condition are reviewed (e.g., educational applications, treatment possibilities for developmental and psychopathological disorders, the development of neural prostheses). All of this foreshadows the potential societal implications of the pragmatic turn. The chapter concludes that an action-oriented approach emphasizes a continuum of interaction between technical aspects of cognitive systems and robotics, biology, psychology, the social sciences, and the humanities, where the individual is part of a grounded cultural system.

show abstract

Section: Action In Robotic Perceptionmentioning

confidence: 92%

Implications of Action-Oriented Paradigm Shifts in Cognitive Science

et al. 2016

View full text Add to dashboard Cite

show abstract

“…We believe that such questions can not always be resolved from single images alone and are exploring methods for active object hypothesis testing and validation [75]. We also use the presented system to learn more about objects through tactile exploration of object shape [76] and study objects in terms of object-related sensorimotor contingencies [77].…”

Section: Discussionmentioning

confidence: 99%

Detecting, segmenting and tracking unknown objects using multi-label MRF inference

Björkman

Bergström

Kragić

2014

Computer Vision and Image Understanding

Self Cite

View full text Add to dashboard Cite

This article presents a unified framework for detecting, segmenting and tracking unknown objects in everyday scenes, allowing for inspection of object hypotheses during interaction over time. A heterogeneous scene representation is proposed, with background regions modeled as a combinations of planar surfaces and uniform clutter, and foreground objects as 3D ellipsoids. Recent energy minimization methods based on loopy belief propagation, tree-reweighted message passing and graph cuts are studied for the purpose of multi-object segmentation and benchmarked in terms of segmentation quality, as well as computational speed and how easily methods can be adapted for parallel processing. One conclusion is that the choice of energy minimization method is less important than the way scenes are modeled. Proximities are more valuable for segmentation than similarity in colors, while the benefit of 3D information is limited. It is also shown through practical experiments that, with implementations on GPUs, multiobject segmentation and tracking using state-of-art MRF inference methods is feasible, despite the computational costs typically associated with such methods.

show abstract

“…Assuming that sensory information is always measured with respect to an object reference, as done in [30], we reformulate the function (2) as follows:…”

Section: A Action-effect As Forward Modelmentioning

confidence: 99%

“…In the experiments we kept the height fixed to positions along the lower part of the object, such as the objects do not fall, and applied actions in directions parallel to the support where the object is placed. The velocity a V extends the dimensionality of the action space previously defined by a = (a X ) in [30]. Some object properties such as rolling for the balls, are only "triggered" from a given velocity.…”

Section: A Model For Pushingmentioning

confidence: 99%

“…Many studies on action-grounded learning rely on a specific controller and describe actions as high-level behaviors without specifying their parameters [10], [12], [19]. In [30] we defined a pushing action by the contact position on the one-dimensional horizontal axis. In this study we consider a multi-dimensional action space incorporating the velocity of the push and deal with a larger set of objects.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Sensorimotor Learning Framework for Object Categorization

Högman

Björkman

Maki

et al. 2016

IEEE Trans. Cogn. Dev. Syst.

Self Cite

View full text Add to dashboard Cite

Abstract-This paper presents a framework that enables a robot to discover various object categories through interaction. The categories are described using action-effect relations, i.e. sensorimotor contingencies rather than more static shape or appearance representation. The framework provides a functionality to classify objects and the resulting categories, associating a class with a specific module. We demonstrate the performance of the framework by studying a pushing behavior in robots, encoding the sensorimotor contingencies and their predictability with Gaussian Processes. We show how entropy-based action selection can improve object classification and how functional categories emerge from the similarities of effects observed among the objects. We also show how a multidimensional action space can be realized by parameterizing pushing using both position and velocity.

show abstract

Interactive object classification using sensorimotor contingencies

Cited by 15 publications

References 27 publications

Implications of Action-Oriented Paradigm Shifts in Cognitive Science

Implications of Action-Oriented Paradigm Shifts in Cognitive Science

Detecting, segmenting and tracking unknown objects using multi-label MRF inference

A Sensorimotor Learning Framework for Object Categorization

Contact Info

Product

Resources

About