Intelligent behavior in humans is largely associated with encoding information to discrete symbols. However, symbolic behavior in robotic systems is not widespread, mainly due to lack of tools that constitute symbol-based automation implementable in real time. This thesis proposes real-time algorithms that facilitate and promote symbol-based action and sensing. In particular, we show that time-efficient control with symbolic actions is possible, using mode scheduling and, subsequently, finite state machines. Serving as the link between action and sensing is a model predictive control algorithm that performs adaptive exploration, driven by an information distribution. This exploration strategy is the key that enables sensing---i.e., learning and tracking---symbols in a variety of settings, reinforcing the importance of information equivalence. The proposed methodologies are validated through simulation examples, reflecting real-world situations, as well as experimentation that verifies real-time execution and implementability. In the final example of this thesis, we demonstrate---in experiment, using a mobile robot performing tactile exploration---how learned symbols can be subsequently employed as self-localization landmarks through their information signature.

Last modified

• 10/22/2018

Creator

• Anastasia Mavrommati

DOI

• https://doi.org/10.21985/N28458

Subject

• Mechanical Engineering

Keyword

• Robotics

Date created

• 2017-01-01

Resource type

• Dissertation

Rights statement

• In Copyright