Learning to Look: A Dynamic Neural Fields Architecture for Gaze Shift Generation

Bell, Christian; Storck, Tobias; Sandamirskaya, Yulia

doi:10.1007/978-3-319-11179-7_88

Cited by 5 publications

(6 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…E.g., the vector-integration to end-point (VITE) neuronal motor-control model generates movement based on the currently estimated pose and the stored goal pose (Grossberg, 1988). Alternatively, one can use a saccadic eye-movement-generating neuronal architecture (Bell et al, 2014;Storck, 2014, 2015) to initiate the gaze to the memorized pose.…”

Section: Representing the Visual Scene In The Network (Map Formation)mentioning

confidence: 99%

An On-chip Spiking Neural Network for Estimation of the Head Pose of the iCub Robot

et al. 2020

Self Cite

View full text Add to dashboard Cite

In this work, we present a neuromorphic architecture for head pose estimation and scene representation for the humanoid iCub robot. The spiking neuronal network is fully realized in Intel's neuromorphic research chip, Loihi, and precisely integrates the issued motor commands to estimate the iCub's head pose in a neuronal path-integration process. The neuromorphic vision system of the iCub is used to correct for drift in the pose estimation. Positions of objects in front of the robot are memorized using on-chip synaptic plasticity. We present real-time robotic experiments using 2 degrees of freedom (DoF) of the robot's head and show precise path integration, visual reset, and object position learning on-chip. We discuss the requirements for integrating the robotic system and neuromorphic hardware with current technologies.

show abstract

Section: Representing the Visual Scene In The Network (Map Formation)mentioning

confidence: 99%

An On-chip Spiking Neural Network for Estimation of the Head Pose of the iCub Robot

et al. 2020

Self Cite

View full text Add to dashboard Cite

show abstract

“…ballistic) camera movements toward objects (here, the salient portions of the visual input). This adaptive looking system has been introduced recently [11], [12], [10] and is described briefly in this paper for completeness in Section III-A. The looking module generates motor commands that drive the motors of the camera head (the pink solid line in the figure), and at the same time creates a representation of the selected object in the gaze-direction space, i.e.…”

Section: A Dynamic Neural Fieldsmentioning

confidence: 99%

“…We have demonstrated how the involved sensorimotor mapping between the retinal positions of the selected objects and the motor commands, needed to saccade toward these objects may be initially learned and constantly updated [9], [10], [11]. We have implemented the architecture to control an autonomous humanoid robot to show that the model indeed may generate behaviour and learn autonomously in a real-world setting [12].…”

Section: Introductionmentioning

confidence: 99%

Learning to reach after learning to look: A study of autonomy in learning sensorimotor transformations

Rudolph¹,

Storck²,

Sandamirskaya

2015

2015 International Joint Conference on Neural Networks (IJCNN)

Self Cite

View full text Add to dashboard Cite

Abstract-This paper presents a neurally-inspired architecture for learning to reach toward visually-perceived targets. The whole behavioural loop from object perception to motor control is realised in the architecture of interconnected Dynamic Neural Fields. The sensory-motor mappings, involved in generation of saccadic gaze shifts and reaching arm movements, adapt in the system autonomously along with the generated behaviour. A network of neural-dynamic nodes organises activation and deactivation of the behavioural modules of the architecture in time, leading to an autonomous process model of development of looking and reaching. The architecture was implemented and validated in a simulated robotic agent.

show abstract

“…Recently, we have demonstrated how the presented architecture is capable of learning to perform precise saccades and to adapt to changes in the environment or in the sensorimotor plant [11]. Here, we modified the learning processes by initialising the gain maps with small random numbers and simulating a more natural learning process, in which the maps are learned in a less controlled learning session.…”

Section: A Gain Maps Learningmentioning

confidence: 99%

“…We have demonstrated the basic functionality of this model recently [11]. Here, we focus on the capability of the network to predict the motor outcome of the potential saccade, to use this prediction to create memories in a motor-based (non-retinal) reference frame, and, finally, to perform a sequence of saccades from memory.…”

Section: Introductionmentioning

confidence: 99%

Neural-dynamic architecture for looking: Shift from visual to motor target representation for memory saccades

Sandamirskaya

Storck

2014

4th International Conference on Development and Learning and on Epigenetic Robotics

Self Cite

View full text Add to dashboard Cite

Looking at objects is one of the most basic sensorimotor behaviours, which requires calibration of the perceptual and motor systems. Recently, we have introduced a neural-dynamic architecture, in which the sensorimotor transformations, which lead to precise saccadic gaze shifts, is initially learned and is autonomously updated if changes in the environment or in the motor plant of the agent require adaptation. Here, we demonstrate how the allocentric, gazedirection independent memory representations may be formed in this architecture and how sequences of precise gaze shifts may be generated to memorised targets. Our simulated robotic experiments demonstrate functioning of the architecture on an autonomous embodied agent.

show abstract

Learning to Look: A Dynamic Neural Fields Architecture for Gaze Shift Generation

Cited by 5 publications

References 19 publications

An On-chip Spiking Neural Network for Estimation of the Head Pose of the iCub Robot

An On-chip Spiking Neural Network for Estimation of the Head Pose of the iCub Robot

Learning to reach after learning to look: A study of autonomy in learning sensorimotor transformations

Neural-dynamic architecture for looking: Shift from visual to motor target representation for memory saccades

Contact Info

Product

Resources

About