The paper demonstrates that aspects of resilience of supervisory Cyber-Physical Systems (CPSs) can be improved through the inclusion of appropriate learning modules in the subordinate autonomous agents. During normal operation, individual agents keep track of their supervisor's commands and utilize the learning module, based on Grammatical Inference, to learn aspects of the organizational structure of the general system and role assignments. It is shown that in cases that the supervisor fails or communication to subordinates is disrupted, these agents are able to recover normalcy of operations. Guaranteeing normalcy recovery in supervisory CPSs is critical in cases of a catastrophic failure or malicious attack.