Auton design

[thefangbear]

Opening this up as a discussion thread for autonomous mode design.

My perspective:

• Use recording-based approach
• Time the HIDs because timing each subsystem is costly and inaccurate
• All recording must be done on one, single thread (the same thread). No concurrency can be involved
• Recording must be event-based
• All HID must implement an abstract class, which has a get() public abstract method that has a generic return type. The get() abstract method does two things: 1) call and memoize the result of another private abstract method, let's say, rawGet() 2) return the result We serialize the memoized array list upon destruct.
• the generic semantic ensures that we can implement a wide array of HIDs.
• for recording, record throughout the entire active period, and when playing back, we can customize a key that triggers the playback (i.e. we can go to a certain position, and then begin)
• Count the entire state of the HID device which composes of multiple actors a single state in the recording period. Recording single button presses is not what we want. Upon serialization, we shall serialize based on the notion of a "state".
• Disk write and read must be atomic as well, in terms of time.

[thefangbear]

Placing this on the backlog right now...

Just to add that FPGATimer() must be used to log the time of each command for play back.

When playing back, we must simulate ArrayList writes (also simulate reads when recording) to account for fetch latencies.