Research documentation
ยท
Experimental record
Research
Research question
The central question is whether different self-healing paradigms behave differently when exposed to the same controlled failures in an industrial multi-agent control environment.
Primary question
How do four recovery-mechanism paradigms compare under the same standardized fault suite?
Secondary questions
- Does one recovery paradigm dominate across all fault classes, or is the best mechanism fault-dependent?
- Is there a trade-off between recovery speed and recovery safety?
Future work
- How does recovery overhead scale as simultaneous faults increase? The current sweep injects one fault per episode, so this question is not yet addressed.
Scope
The fault suite contains sensor noise, actuator degradation, agent dropout, communication loss, and Byzantine/adversarial agents. Each fault is represented in permanent and transient forms.