No abstract

Testing is a promising way to gain trust in neural action policies π. Previous work on policy testing in sequential decision making targeted environment behavior leading to failure conditions. But if the failure is unavoidable given that behavior, then π is not actually to blame. For a situation to qualify as a "bug" in π, there must be an alternative policy π' that does better. We introduce a generic policy testing framework based on that intuition. This raises the bug confirmation problem, deciding whether or not a state is a bug. We analyze the use of optimistic and pessimistic bounds for the design of test oracles approximating that problem. We contribute an implementation of our framework in classical planning, experimenting with several test oracles and with random-walk methods generating test states biased to poor policy performance and/or state novelty. We evaluate these techniques on policies π learned with ASNets. We find that they are able to effectively identify bugs in these π, and that our random-walk biases improve over uninformed baselines.

Temporal stream logic (TSL) extends LTL with updates and predicates over arbitrary function terms. This allows for specifying data-intensive systems for which LTL is not expressive enough. In the semantics of TSL, functions and predicates are left uninterpreted. In this paper, we extend TSL with first-order theories, enabling us to specify systems using interpreted functions and predicates such as incrementation or equality. We investigate the satisfiability problem of TSL modulo the standard underlying theory of uninterpreted functions as well as with respect to Presburger arithmetic and the theory of equality: For all three theories, TSL satisfiability is neither semi-decidable nor co-semi-decidable. Nevertheless, we identify three fragments of TSL for which the satisfiability problem is (semi-)decidable in the theory of uninterpreted functions. Despite the undecidability, we present an algorithm – which is not guaranteed to terminate – for checking the satisfiability of a TSL formula in the theory of uninterpreted functions and evaluate it: It scales well and is able to validate assumptions in a real-world system design.

Automatic synthesis from temporal logic specifications is an attractive alternative to manual system design, due to its ability to generate correct-by-construction implementations from high-level specifications. Due to the high complexity of the synthesis problem, significant research efforts have been directed at developing practically efficient approaches for restricted specification language fragments. In this paper we focus on the Safety LTL fragment of Linear Temporal Logic (LTL) syntactically extended with bounded temporal operators. We propose a new synthesis approach with the primary motivation to solve efficiently the synthesis problem for specifications with bounded temporal operators, in particular those with large bounds. The experimental evaluation of our method shows that for this type of specifications it outperforms state-ofart synthesis tools, demonstrating that it is a promising approach to efficiently treating quantitative timing constraints in safety specifications.