Planning with Trust for Human-Robot Collaboration

Abstract

Perspective-taking is the ability to perceive or understand a situation or concept from another individual's point of view, and is crucial in daily human interactions. Enabling robots to perform perspective-taking remains an unsolved problem; existing approaches that use deterministic or handcrafted methods are unable to accurately account for uncertainty in partially-observable settings. This work proposes to address this limitation via a deep world model that enables a robot to perform both perception and conceptual perspective taking, i.e., the robot is able to infer what a human sees and believes. The key innovation is a decomposed multi-modal latent state space model able to generate and augment fictitious observations/emissions. Optimizing the ELBO that arises from this probabilistic graphical model enables the learning of uncertainty in latent space, which facilitates uncertainty estimation from high-dimensional observations. We tasked our model to predict human observations and beliefs on three partially-observable HRI tasks. Experiments show that our method significantly outperforms existing baselines and is able to infer visual observations available to other agent and their internal beliefs.

Robot Capability and Intention in Trust-Based Decisions Across Tasks

“…Misplaced trust in robots can lead to poor task-allocation and unsatisfactory outcomes. In recognition of its importance, there has been a concerted effort in the research community to better understand the formation and dynamics of trust in robots and automation [4]- [10].…”

“…Nevertheless, there remains crucial gaps in our understanding of human-robot trust, particularly in the role of inferred robot "intention", i.e., what people believe the robot is trying to achieve. Prior research has focussed largely on inferred capability [7], [9], [10], which has been shown to be a primary antecedent to trust [11], [12]. However, with the advancement of robot technology (e.g., artificial intelligence), robots are increasingly poised to achieve peer-like collaboration with humans.…”

Robots that Take Advantage of Human Trust

“…Most relevant here is work by Chen et al [6], who treat the human's trust as a latent variable, and integrate trust into the robot's decision making. These previous works [3]- [6] consider cases where the human and collaborative robot share a common objective and the robot should maximize trust; by contrast, we are interested in robots that leverage trust to influence users.…”

“…Here, we revisit the two-player game outlined in Section III-A, and use the trusting human model (6) to reformulate this game as an instance of a single-agent Markov decision process (MDP). Expressing the trusting human model as an MDP enables us to leverage existing methods to solve for u * ,τ r .…”

“…We compared three formulations for selecting robot behavior: conservatively following the Nash equilibrium (Nash), optimistically solving the two-player game (Optimistic), or computing optimal behavior for a trusting human (Trust). For Nash the robot follows (3), for Optimistic the robot solves (4), and for Trust the robot leverages (6). Each of these robot strategies makes a different assumption about the human user: Nash acts as if the human knew θ r , Optimistic assumes the human believes the robot objective isθ r , and Trust simplifies the rational human model.…”