Following Spinrad, a robust algorithm A for a graph class C and an 
algorithmic problem Pi works as follows: Given the input graph G,
if G is in C then A solves the problem Pi correctly,
if $G$ is not in C then A either solves Pi correctly 
or gives a proof that G is not in C.

Thus, a robust algorithm produces a ``correct'' result in any case, 
which may be either a correct solution to the problem or the answer 
that the input graph is not in the class. 
This is of crucial importance whenever the recognition 
time bound for C is worse than solving the problem on the graph 
class.

The talk will survey and discuss robust algorithms for standard algorithmic
graph problems such as Clique, Independent Set and Coloring on particular
graph classes.