| Title | : | Planning Stories with Agent-updates in Dynamic Epistemic Logic |
| Speaker | : | Shikha Singh (IITM) |
| Details | : | Thu, 16 Nov, 2023 3:00 PM @ Google Meet |
| Abstract: | : | This thesis studies modeling and representation of beliefs in a multiagent scenario, with focus on deception. We look at how false beliefs of selected agents can be represented in Kripke models, especially when agents are adde d or removed in the domain. We look at a well known children’s fable as a running example, in which a clever mouse invents a scary friend, Gruffalo, to scare a predator away. We use Dynamic Epistemic Logic (Van Ditmarsch et al., 2008) to analyze the lies. Lying and deception have been studied earlier in DE L (Van Ditmarsch et al., 2012) and using Theory of Mind (Sakama, 2015; Sarkadi et al., 2019). DE L has been used by the planning community to augment classical planning problems with Knowledge and Beliefs of agents and their epistemic goals (Bolander and Andersen, 2011; Bolander, 2017; Baral et al., 2022). We introduce agent-update operators for selective agent addition and deletion, both honest and deceptive, which modify the Kripke model to ascribe new beliefs and drop the discarded beliefs of agents. This is formulated with a new logic DEL∗ which extends the usual product-update semantics of DE L with a sum-product update. The thesis provides soundness and completeness results for the proposed logic and provides algorithms for model checking and satisfiability of the logic. Next we propose an agent-update logic AUL which extends the usual syntax of DEL with specific agent-update modalities. We provide the theoretical results for the same. A detailed modelling of examples from The Gruffalo (Donaldson, 1999) and its sequel The Gruffalo’s Child (Donaldson, 2004) shows the applicability of our proposed logic in multi-agent epistemic planning. We take a DE L-based epistemic planning approach.We incorporate the sum-product update in a planning algorithm, and formulate planning operators, including deceptive utterances, specialised for predator-prey domains. The traditional framework is extended to allow broadening of the planning domain by adding prey and predators to synthesize plans in an open world where the set of actions and characters may be large. The idea is to illustrate how a deceptive plan (based on agent-updates) could be generated. Web Conference Link : https://meet.google.com/oav-tyth-nhr |
