Deductive approach
	• Start from the final reward model. • Do a reward regression in terms of abstract states itself using a situation calculus framework. • Extremely slow – requires a need for a theorem prover.