An Epistemic Human-Aware Task Planner which Anticipates Human Beliefs and Decisions

TL;DR

The paper tackles planning under belief divergence between humans and robots when execution experiences are disrupted. It introduces an AND/OR search-based offline planner that integrates knowledge reasoning, perspective taking, and situation assessment within a Dynamic Epistemic Logic–based epistemic planning framework to anticipate human beliefs and selectively communicate. It extends prior human-aware planning work to handle higher-order beliefs (up to level two) and explicit communication actions (ask-$p$ and inform-$p$) to align decompositions across possible worlds. The authors demonstrate the approach in two domains, analyze runtime scaling with the number of possible worlds $|W|$ and the non-shared-action bound $K$, and show how the framework reduces ambiguity and improves plan robustness in HRC settings.

Abstract

We present a substantial extension of our Human-Aware Task Planning framework, tailored for scenarios with intermittent shared execution experiences and significant belief divergence between humans and robots, particularly due to the uncontrollable nature of humans. Our objective is to build a robot policy that accounts for uncontrollable human behaviors, thus enabling the anticipation of possible advancements achieved by the robot when the execution is not shared, e.g. when humans are briefly absent from the shared environment to complete a subtask. But, this anticipation is considered from the perspective of humans who have access to an estimated model for the robot. To this end, we propose a novel planning framework and build a solver based on AND-OR search, which integrates knowledge reasoning, including situation assessment by perspective taking. Our approach dynamically models and manages the expansion and contraction of potential advances while precisely keeping track of when (and when not) agents share the task execution experience. The planner systematically assesses the situation and ignores worlds that it has reason to think are impossible for humans. Overall, our new solver can estimate the distinct beliefs of the human and the robot along potential courses of action, enabling the synthesis of plans where the robot selects the right moment for communication, i.e. informing, or replying to an inquiry, or defers ontic actions until the execution experiences can be shared. Preliminary experiments in two domains, one novel and one adapted, demonstrate the effectiveness of the framework.

Figures (3)

• Figure 1: Our planning framework is endowed with the ability to make the difference between H&R shared and individual execution experiences in the planned activities. It can anticipate potential belief divergence between H&R and also estimate the updated beliefs of H when they meet again (situation assessment (SA)) based on a distinction between observable and non-observable facts. This will be used to plan communicative actions or adapt the R's plan to ensure the shared experience of some actions. In this diagram, we roughly depict what happens when H&R no longer share the execution experience, H has independent experience (blue), while R progresses towards the goal (green), with anticipated traces (in gray) depicting other estimated courses of action that the robot can choose along with the green trace but from the H's perspective. Upon co-presence at place, SA eliminates impossible worlds, e.g., those with state property prop4=F (since it is observable), aiding H to ignore wrongly estimated worlds.
• Figure 2: Three cubes $c_r$ (red), $c_y$ (yellow), and $c_w$ (white) are shown. $c_r$ and $c_y$ are placed on mt (main table), and $c_w$ is on ot (other table). There are two boxes, $box_1$ and $box_2$, placed on mt, which can be either transparent or opaque. The shared task is to organize the cubes in a way that cubes from one table are placed in one box. The choice of which box is flexible as long as each table's cubes end up in separate boxes.
• Figure 3: We represent a state $(s_i)$, action $(a_i)$, and how applying $a_i$ in $s_i$ leads to next state $(s_{i+1} = s_{i} \otimes a_{i})$. $f$ is a formula that captures if H&R were co-present when the events took place. Common facts for both worlds, such as opaque($box_1$), are not shown. Also, each world is fully defined, with either an atom or its negation holding true.