Jörg Hoffmann

Katharina Stein, Nils Hodel, Daniel Fišer et al.

LLMs have recently been used to generate Python programs representing generalized plans in PDDL planning, i.e., plans that generalize across the tasks of a given PDDL domain. Previous work proposed a framework consisting of three steps: the LLM first generates a summary and then a strategy for the domain, both in natural language, and then implements that strategy as a Python program, that gets debugged on example planning tasks. In that work, only one strategy is generated and passed directly to the program generation. If the strategy is incorrect, its implementation will therefore result in an incorrect generalized plan. Here, we introduce an approach that generates the strategy in the form of pseudocode and enables automatic debugging of the pseudocode, hence allowing us to identify and fix errors prior to the generation of the generalized plan itself. Additionally, we extend the Python debugging phase with a reflection step prompting the LLM to pinpoint the reason for the observed plan failure. Finally, we take inspiration from LLM code generation to produce several program variants and pick the best one. Running experiments on 17 benchmark domains, we show that these extensions substantially improve (and never deteriorate) the quality of the generalized plans. In 12 of the domains, our best Python programs solve all tasks that can be generated with the respective instance generator.

Rebecca Eifler, Jörg Hoffmann

In a variety of application settings, the user preference for a planning task - the precise optimization objective - is difficult to elicit. One possible remedy is planning as an iterative process, allowing the user to iteratively refine and modify example plans. A key step to support such a process are explanations, answering user questions about the current plan. In particular, a relevant kind of question is "Why does the plan you suggest not satisfy $p$?", where p is a plan property desirable to the user. Note that such a question pertains to plan space, i.e., the set of possible alternative plans. Adopting the recent approach to answer such questions in terms of plan-property dependencies, here we implement a tool and user interface for human-guided iterative planning including plan-space explanations. The tool runs in standard Web browsers, and provides simple user interfaces for both developers and users. We conduct a first user study, whose outcome indicates the usefulness of plan-property dependency explanations in iterative planning.

Arne Köhn, Julia Wichlacz, Álvaro Torralba et al.

When generating technical instructions, it is often convenient to describe complex objects in the world at different levels of abstraction. A novice user might need an object explained piece by piece, while for an expert, talking about the complex object (e.g. a wall or railing) directly may be more succinct and efficient. We show how to generate building instructions at different levels of abstraction in Minecraft. We introduce the use of hierarchical planning to this end, a method from AI planning which can capture the structure of complex objects neatly. A crowdsourcing evaluation shows that the choice of abstraction level matters to users, and that an abstraction strategy which balances low-level and high-level object descriptions compares favorably to ones which don't.

Timo P. Gros, Daniel Höller, Jörg Hoffmann et al.

Learning-based approaches for solving large sequential decision making problems have become popular in recent years. The resulting agents perform differently and their characteristics depend on those of the underlying learning approach. Here, we consider a benchmark planning problem from the reinforcement learning domain, the Racetrack, to investigate the properties of agents derived from different deep (reinforcement) learning approaches. We compare the performance of deep supervised learning, in particular imitation learning, to reinforcement learning for the Racetrack model. We find that imitation learning yields agents that follow more risky paths. In contrast, the decisions of deep reinforcement learning are more foresighted, i.e., avoid states in which fatal decisions are more likely. Our evaluations show that for this sequential decision making problem, deep reinforcement learning performs best in many aspects even though for imitation learning optimal decisions are considered.

Patrick Speicher, Marcel Steinmetz, Jörg Hoffmann et al.

Penetration testing is a well-established practical concept for the identification of potentially exploitable security weaknesses and an important component of a security audit. Providing a holistic security assessment for networks consisting of several hundreds hosts is hardly feasible though without some sort of mechanization. Mitigation, prioritizing counter-measures subject to a given budget, currently lacks a solid theoretical understanding and is hence more art than science. In this work, we propose the first approach for conducting comprehensive what-if analyses in order to reason about mitigation in a conceptually well-founded manner. To evaluate and compare mitigation strategies, we use simulated penetration testing, i.e., automated attack-finding, based on a network model to which a subset of a given set of mitigation actions, e.g., changes to the network topology, system updates, configuration changes etc. is applied. Using Stackelberg planning, we determine optimal combinations that minimize the maximal attacker success (similar to a Stackelberg game), and thus provide a well-founded basis for a holistic mitigation strategy. We show that these Stackelberg planning models can largely be derived from network scan, public vulnerability databases and manual inspection with various degrees of automation and detail, and we simulate mitigation analysis on networks of different size and vulnerability.

Joerg Hoffman, Ingo Weber, Frank Michael Kraft

Planning is concerned with the automated solution of action sequencing problems described in declarative languages giving the action preconditions and effects. One important application area for such technology is the creation of new processes in Business Process Management (BPM), which is essential in an ever more dynamic business environment. A major obstacle for the application of Planning in this area lies in the modeling. Obtaining a suitable model to plan with -- ideally a description in PDDL, the most commonly used planning language -- is often prohibitively complicated and/or costly. Our core observation in this work is that this problem can be ameliorated by leveraging synergies with model-based software development. Our application at SAP, one of the leading vendors of enterprise software, demonstrates that even one-to-one model re-use is possible. The model in question is called Status and Action Management (SAM). It describes the behavior of Business Objects (BO), i.e., large-scale data structures, at a level of abstraction corresponding to the language of business experts. SAM covers more than 400 kinds of BOs, each of which is described in terms of a set of status variables and how their values are required for, and affected by, processing steps (actions) that are atomic from a business perspective. SAM was developed by SAP as part of a major model-based software engineering effort. We show herein that one can use this same model for planning, thus obtaining a BPM planning application that incurs no modeling overhead at all. We compile SAM into a variant of PDDL, and adapt an off-the-shelf planner to solve this kind of problem. Thanks to the resulting technology, business experts may create new processes simply by specifying the desired behavior in terms of status variable value changes: effectively, by describing the process in their own language.