Planning and Acting with Non-Deterministic Events: Navigating between Safe States

Abstract

Automated Planning addresses the problem of finding a sequence of actions, a plan, transforming the environment from its initial state to some goal state. In real-world environments, exogenous events might occur and might modify the environment without agent's consent. Besides disrupting agent's plan, events might hinder agent's pursuit towards its goals and even cause damage (e.g. destroying the robot).In this paper, we leverage the notion of Safe States in dynamic environments under presence of non-deterministic exogenous events that might eventually cause dead-ends (e.g. “damage” the agent) if the agent is not careful while executing its plan. We introduce a technique for generating plans that constrains the number of consecutive “unsafe” actions in a plan and a technique for generating “robust” plans that effectively evade event effects. Combination of both approaches plans and executes robust plans between safe states. We empirically show that such an approach effectively navigates the agent towards its goals in spite of presence of dead-ends.

On Eventual Applicability of Plans in Dynamic Environments with Cyclic Phenomena

“…Addressing the problem by relaxing non-deterministic events, i.e., by leveraging classical planning techniques such as FF-replan (Yoon, Fern, and Givan 2007), where if the agent is in an unexpected state it re-plans, is efficient but prone to dead-ends. Reasoning about "dangerous states" improves the success rate of FF-replan-like approaches, yet it does not guarantee avoiding dead-ends (Chrpa, Gemrot, and Pilát 2017;Chrpa, Gemrot, and Pilát 2020).…”

“…To tackle the dead-end proneness, one has to enhance classical planning (and re-planning) techniques by a mechanism that prohibits the agent to get into a dead-end state in spite of event occurrence. Chrpa, Gemrot, and Pilát (2020) adapted the notion of safe states (Cserna et al 2018) for planning tasks with non-deterministic events, where a state is safe if no sequence of events can transform it to a dead-end state. If such an event sequence exists, the state is unsafe.…”

“…If such an event sequence exists, the state is unsafe. The technique proposed by Chrpa, Gemrot, and Pilát (2020) generates a (reference) plan while minimising the number of consecutive unsafe actions since, following the rule of thumb, the shorter the sequence of unsafe states is the higher chance to safely passing it is. The (reference) plan is then followed such that unsafe states can only be traversed by robust plans connecting one safe state to another.…”

“…If there is no way of transiting an unsafe area via a robust plan, the agent gets stuck forever (albeit in a safe state). Hence, the technique of Chrpa, Gemrot, and Pilát (2020) works only on a subclass of planning tasks and it might not be known up front whether the agent has a chance to reach its goal, or whether it gets stuck somewhere.…”

“…We introduce techniques for analysing the problem and determining reversible and possibly irreversible events (possibly unsafe ones), that is, identifying cyclic phenomena and possible consequences of agent's interference with them. Then, we show under which conditions a state is safe, so safe states do not have to be specified on top of domain and problem specification (as required by the technique of Chrpa, Gemrot, and Pilát (2020)). We also show under which conditions it is possible that a sequence of actions traversing through unsafe states, denoted as an unsafe bridge, might eventually become a robust plan and hence safely applicable.…”