Visual Target Sequence Prediction via Hierarchical Temporal Memory Implemented on the iCub Robot
In this article, we present our initial work on sequence prediction of a visual target by implementing a cortically inspired method, namely Hierarchical Temporal Memory (HTM). As a preliminary test, we employ HTM on periodic functions to quantify predicti
- PDF / 1,132,256 Bytes
- 12 Pages / 439.37 x 666.142 pts Page_size
- 85 Downloads / 177 Views
Abstract. In this article, we present our initial work on sequence prediction of a visual target by implementing a cortically inspired method, namely Hierarchical Temporal Memory (HTM). As a preliminary test, we employ HTM on periodic functions to quantify prediction performance with respect to prediction steps. We then perform simulation experiments on the iCub humanoid robot simulated in the Neurorobotics Platform. We use the robot as embodied agent which enables HTM to receive sequences of visual target position from its camera in order to predict target positions in different trajectories such as horizontal, vertical and sinusoidal. The obtained results indicate that HTM based method can be customized for robotics applications that require adaptation of spatiotemporal changes in the environment and acting accordingly. Keywords: Hierarchical Temporal Memory Neurorobotics Platform
1
·
Sequence prediction
·
Introduction
Understanding mammalian brain functions and mechanisms have been attractive research field for robotics and artificial intelligence (AI). These functions and mechanisms provide proof of concept solutions for known problems in robotics and AI such as operating massively parallel processes with relatively low power consumption, processing noisy sensory information for action execution, decision making, etc. Thus, brain-inspired approaches have been investigated to derive principles of brain functions and mechanisms in order to propose solutions for various existing challenges. The several known examples are image recognition in large databases by means of deep neural networks [1], artificial neural network based path planning [2], to mention a few. Although these methods perform well in a specific task with notable accuracy, they lack in biological plausibility and generalization of the solution in different domains. To build more biologically realistic models with generalization capabilities, a method based on operating principles of Neocortex has been proposed in [3] and c Springer International Publishing Switzerland 2016 N.F. Lepora et al. (Eds.): Living Machines 2016, LNAI 9793, pp. 119–130, 2016. DOI: 10.1007/978-3-319-42417-0 12
120
M. Kirtay et al.
named as Hierarchical Temporal Memory (HTM). HTM based solutions1 have been successfully implemented in a considerable number of applications and in a wide range of areas where sequential learning and generative prediction aimed at anomaly detection, image classification, rogue behavior detection, geospatial tracking. In robotics HTM can be used to anticipate the world future states in order to properly control motion and deal with the continuously changing environment. Humans appear to solve this problem by predicting the changes in their sensory system as a consequence of their actions [5]. The predictions are obtained using internal models that represent their own bodies and the external objects dynamics [6–8]. There are three main types of internal models [7]: the forward models, the environment models, and the inverse models. The forward mo
Data Loading...
