Improving object learning through manipulation and robot self-identification

Abstract

Abstract-We present a developmental approach that allows a humanoid robot to continuously and incrementally learn entities through interaction with a human partner in a first stage before categorizing these entities into objects, humans or robot parts and using this knowledge to improve objects models by manipulation in a second stage. This approach does not require prior knowledge about the appearance of the robot, the human or the objects. The proposed perceptual system segments the visual space into proto-objects, analyses their appearance, and associates them with physical entities. Entities are then classified based on the mutual information with proprioception and on motion statistics. The ability to discriminate between the robot's parts and a manipulated object then allows to update the object model with newly observed object views during manipulation. We evaluate our system on an iCub robot, showing the independence of the self-identification method on the robot's hands appearances by wearing different colored gloves. The interactive object learning using self-identification shows an improvement in the objects recognition accuracy with respect to learning through observation only.

40 years of cognitive architectures: core cognitive abilities and practical applications

“…Only a handful of architectures implement suppression mechanisms. For instance, suppression of task irrelevant visual information is done in only three architectures: in BECCA irrelevant features are suppressed through the WTA mechanisms [437], in DSO background features are suppressed to speed up processing [610], in MACSi depth is used to ignore regions that are not reachable by the robot [337] and in ARCADIA non-central regions of the visual field are inhibited at each cycle since the cue always appears in the center [72]. Another suppression mechanism is inhibition of return (IOR), which prevents the visual system from attending to the same salient stimuli.…”

Time-Lapse Light Field Photography for Perceiving Non-Lambertian Scenes

“…A number of approaches use robots to acquire 3D structure of objects either by moving the camera or moving the object [12,1,9,28,24,10,16,15,11,1]. Our approach, instead, uses a model based around light fields, incorporating both intensity and direction of light rays.…”

Object Learning Through Active Exploration

“…For example, we demonstrated how the robot can learn autonomously its visuomotor representations in simple visual servoing tasks [23], and to recognize objects from observation and interaction [12], [33]. • Physical interaction with the environment: Intelligence requires the interplay between the human baby with his surrounding, i.e., people and objects.…”

“…It uses minimal prior knowledge of the environment: in particular it is able to incrementally learn robot, caregiver hands and object appearance during interaction with caregivers and objects without complementary supervision. The system has been described in details in [12], [54]. A short overview is given here to complement the architecture presentation.…”

“…The proprioceptive sensors, the inertial and the proximal force/torque sensors embedded on the platform, combined with its kinematics and dynamics modeling, are fed to numerous modules of the architecture: modules for controlling gaze, posture and reaching movements [65], modules for controlling the whole-body dynamics [35], the robot compliance and its contact forces [36], modules for learning incrementally the visuomotor models of the robot [23], the vision modules recognizing the robot's self-body in the visual space [33], the basic modules for speech and gaze tracking [24], just to cite a few. 5 The architecture is implemented as a set of concurrent modules exchanging information with the robot thanks to the YARP middleware [66].…”