Abstract

Commodity computers continue to become ever faster, through increases in clock speed and now through addition of multiple processor cores. The ability to use these resources in a closed-loop scientific or engineering task enables new types of applications with greatly reduced needs on specialized software and hardware. However, pre-emptive multitasking operating systems, which are now nearly ubiquitous due to many practical benefits, present a serious challenge to those who would employ computers in realtime tasks. A program's control of the CPU is non-deterministic. Associated with this is a perhaps-unexpected benefit: high-level languages such as Python, which were previously not useful for use in realtime systems due to non-deterministic features such as automatic garbage collection, now present virtually no additional complication beyond that imposed by a modern operating system. Here I will discuss three examples of real-time computational tasks from my own research and that of colleagues in which we dealt with these issues at several timescales. First is a hard realtime motor control system which uses RTAI/Linux to acheive sub-millisecond precision for analog input and motor command output. Second is the Vision Egg, a software library for display of complex graphics in realtime for visual neuroscience applications. Finally, I present an experimental apparatus which tracks the position and orientation freely flying animals in realtime using multiple synchronized digital cameras. The issues of process scheduling, file IO, multi-threading, interfacing with hardware, presenting a user interface will be discussed in each of these examples.

You may download the slides from the talk.