Unsupervised Emergence of Egocentric Spatial Structure from Sensorimotor Prediction
Laflaquiere, A. & Garcia Ortiz, M. ORCID: 0000-0003-4729-7457 (2019). Unsupervised Emergence of Egocentric Spatial Structure from Sensorimotor Prediction. Advances in Neural Information Processing Systems 32 (NIPS 2019),
Abstract
Despite its omnipresence in robotics application, the nature of spatial knowledgeand the mechanisms that underlie its emergence in autonomous agents are stillpoorly understood. Recent theoretical works suggest that the Euclidean structure ofspace induces invariants in an agent’s raw sensorimotor experience. We hypothesizethat capturing these invariants is beneficial for sensorimotor prediction and that,under certain exploratory conditions, a motor representation capturing the structureof the external space should emerge as a byproduct of learning to predict futuresensory experiences. We propose a simple sensorimotor predictive scheme, applyit to different agents and types of exploration, and evaluate the pertinence of thesehypotheses. We show that a naive agent can capture the topology and metricregularity of its sensor’s position in an egocentric spatial frame without any a prioriknowledge, nor extraneous supervision.
Publication Type: | Article |
---|---|
Additional Information: | © The Authors. |
Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science |
Departments: | School of Science & Technology > Computer Science |
Download (1MB) | Preview
Export
Downloads
Downloads per month over past year