Possible research for fault tolerance and robustness in autonomous systems:

1. Agent based architecture: Though a three tier architecture is predominant in autonomous embedded systems, the agent approach could work because of 1.Avoiding semantic mismatch given that all agents have the same symbolic representation; 2.Increasing modularity by separating functionality into agents.
2. Plan analysis: Checking plans (created by decisional mechanisms) on-line.  This would arguably increase the validity and relevance of a plan.
3. Heuristic diversification: Have a pool of heuristics that work inside the decisional mechanism.  All of them can execute at the same time or one can be preferred.
4. Model diversification: Have a pool of implementations and descriptions of the systems.  In such a way that if one fails, the others can be used.
5. Fault forecasting: This refers to the estimation, incidence and consequences of future faults.  There have been few studies in fault forecasting.

They mentioned a simulation and fault injection environment that is not yet complete.  It read like a description of what they are still implementing.