Creating a stable robot agent as a tool for research

	Joseph O'Sullivan  	- Weh7220  	- 3.30pm 10/12/94

The life of an Xavier researcher is akin to that of the systems researcher in the dark ages - the system researcher would think up some nifty whiz-bang ideas for an operating system, compile those changes to get a new kernel, then reboot a machine to test them out. This was found to be a horrible means of working - all those rebootings would cause the machine to be unreliable leading to significant down-time. More annoying, if the modifications didn't work after rebooting, the user didn't have an operating system to fall back upon.

If all the above sounds vaguely familiar to your daily life on Xavier (and it should), take heart. The systems community advanced beyond these dark ages with techniques such as dynamically loadable modules, system monitoring and kernel redundancy. I will discuss how these techniques could be applied to Xavier by creating a basic stable robot agent. This agent would then be responsible for running research level tasks, for maintaining the reliability of the system by monitoring battery and module life, and to its limited capability, for seeking help via speech and zephyr.

I will look forward to discussion about the set of tasks the basic robot agent must be able to perform, thoughts about self monitoring, diagnosis and correction, in addition to opinions about this framework and a proposed implementation.