Using AI

In the LCPF implementation

The LCPF doesn’t use any AI/ML techniques internally, but it might in the future. There are some specific areas where AI/ML might be useful, such as:

• creating “as used” profiles for improving motion effects in ways that are difficult to make “beginner friendly”
• PID tuning
• soft start/stop
• dead reckoning and odometry, especially for Motile projects
• optimizing performance or resource usage
• dynamic tuning of GC constraints and loop timings

During LCPF development

I have been evaluating AI coding assistants while developing the LCPF - primarily CoPilot within VSCode. Personally I’ve found them to be of modest benefit, primarily

• in identifying and re-applying repetitive edits
• fixing typos / syntax issues (although IDE tooling like PyLance already identifies the majority of those without AI) Those alone would offer a valuable improvement in the productivity of my personal workflow, but backing out the time required to cancel and correct the frequently incorrect suggestions significantly reduces the benefit (although I feel it is still a net positive).
• I have made limited use of AI suggested content in writing documentation, primarily when suggested minor edits, rephrasing or summarizations were worthwhile.

However, I’ve been occasionally surprised by the quality of some of the suggestions, and I expect that as these tools improve they will become more useful. I think they might be especially useful for junior and mid range developers working on more mainstream projects.