Sensing with Artificial Tactile Sensors: An Investigation of Spatio-temporal Inference
DOI: 10.1007/978-3-642-23232-9_23 Conference: Towards Autonomous Robotic Systems - 12th Annual Conference, TAROS 2011, Sheffield, UK, August 31 - September 2, 2011. Proceedings
The ease and efficiency with which biological systems deal with several real world problems, that have been persistently challenging
to implement in artificial systems, is a key motivation in biomimetic robotics. In interacting with its environment, the first
challenge any agent faces is to extract meaningful patterns in the inputs from its sensors. This problem of pattern recognition
has been characterized as an inference problem in cortical computation. The work presented here implements the hierarchical
temporal memory (HTM) model of cortical computation using inputs from an array of artificial tactile sensors to recognize
simple Braille patterns. Although the current work has been implemented using a small array of robot whiskers, the architecture
can be extended to larger arrays of sensors of any arbitrary modality.
Available from: Ben Mitchinson
- "In words, where contact of any sort is made, the percept U (n) is that which produces the best match with sensory data, colored by a tendency to stick with the perception of P once it has been made. This two-way ('bottom-up and top-down') approach to generation of a unitary percept from sensory data is closely related to currently popular hierarchical cortical models (see  for a recent review of these in Bayesian terms; see  for an application to shape recognition using whiskers). "
[Show abstract] [Hide abstract]
ABSTRACT: Good performance in unstructured/uncertain environments is an ongoing problem in robotics; in biology, it is an everyday observation. Here, we model a particular biological system—hunting in the Etruscan shrew—as a case study in biomimetic robot design. These shrews strike rapidly and accurately after gathering very limited sensory information from their whiskers; we attempt to mimic this performance by using model-based simultaneous discrimination and localisation of a ‘prey’ robot (i.e. by using strong priors to make sense of limited sensory data), building on our existing low-level models of attention and appetitive behaviour in small mammals. We report performance that is comparable, given the spatial and temporal scale differences, to shrew performance, and discuss what this study reveals about biomimetic robot design in general.
Data provided are for informational purposes only. Although carefully collected, accuracy cannot be guaranteed. The impact factor represents a rough estimation of the journal's impact factor and does not reflect the actual current impact factor. Publisher conditions are provided by RoMEO. Differing provisions from the publisher's actual policy or licence agreement may be applicable.