Vector Symbolic Architectures (VSA) combine a hypervector space and a set of operations on these vectors. Hypervectors provide powerful and noise-robust representations and VSAs are associated with promising theoretical properties for approaching high-level cognitive tasks. However, a major drawback of VSAs is the lack of opportunities to learn them from training data. Their power is merely an effect of good (and elaborate) design rather than learning. We exploit high-level knowledge about the structure of reactive robot problems to learn a VSA based on training data. We demonstrate preliminary results on a simple navigation task. Given a successful demonstration of a navigation run by pairs of sensor input and actuator output, the system learns a single hypervector that encodes this reactive behaviour. When executing (and combining) such VSA-based behaviours, the advantages of hypervectors (i.e. the representational power and robustness to noise) are preserved. Moreover, a particular beauty of this approach is that it can learn encodings for behaviours that have exactly the same form (a hypervector) no matter how complex the sensor input or the behaviours are.
| Identifer | oai:union.ndltd.org:DRESDEN/oai:qucosa.de:bsz:ch1-qucosa-217042 |
| Date | 08 August 2017 |
| Creators | Neubert, Peer, Schubert, Stefan, Protzel, Peter |
| Contributors | TU Chemnitz, Fakultät für Elektrotechnik und Informationstechnik, IEEE, |
| Publisher | Universitätsbibliothek Chemnitz |
| Source Sets | Hochschulschriftenserver (HSSS) der SLUB Dresden |
| Language | English |
| Detected Language | English |
| Type | doc-type:conferenceObject |
| Format | application/pdf, text/plain, application/zip |
| Source | International Conference on Intelligent Robots and Systems (IROS) Workshop on Machine Learning Methods for High-Level Cognitive Capabilities in Robotics, 2016 |
Page generated in 0.0021 seconds