Artificial Intelligence and Machine Learning is increasingly being used in industrial automation systems. Many of these systems are safety-critical, and have hard timing constraints. Here any missed deadlines could have catastrophic consequences. However, real-time AI systems with strict timing deadlines are yet to be carefully investigated in literature. This project will develop a new approach for the design of AI based automation / robotics applications. We will consider deep Neural networks based AI algorithms. To ensure time predictable design, we will introduce a notion of time in these Neural networks inspired by the synchronous programming languages. These algorithms will be implemented using C and will be used in conjunction with the industrial automation standard IEC61499. Techniques will be developed to map these industrial applications for execution on precision timed processor architectures, such as the multicore TCREST platform. A novel application case study in robotics / automation will be developed.
Undergraduate
1) A synchronous approach for the design of deep neural networks
2) Using these algorithms along with BlokIDE tool-chain based on IEC61499: http://pretzel.ece.auckland.ac.nz/#!research?project=iec61499
3) A case study in the robotics / industrial automation lab.
Student will have the freedom to select a suitable case study. This may involve the use of drones and or industrial automation infrastructure.
COMPSYS 303 and 304 are desirable but not compulsory
Lab allocations have not been finalised