Probabilistic Human Intent Recognition for Shared Autonomy in Assistive Robotics

Abstract

Effective human-robot collaboration in shared autonomy requires reasoning about the intentions of the human partner. To provide meaningful assistance, the autonomy has to first correctly predict, or infer, the intended goal of the human collaborator. In this work, we present a mathematical formulation for intent inference during assistive teleoperation under shared autonomy. Our recursive Bayesian filtering approach models and fuses multiple non-verbal observations to probabilistically reason about the intended goal of the user without explicit communication. In addition to contextual observations, we model and incorporate the human agent's behavior as goal-directed actions with adjustable rationality to inform intent recognition. Furthermore, we introduce a user-customized optimization of this adjustable rationality to achieve user personalization. We validate our approach with a human subjects study that evaluates intent inference performance under a variety of goal scenarios and tasks. Importantly, the studies are performed using multiple control interfaces that are typically available to users in the assistive domain, which differ in the continuity and dimensionality of the issued control signals. The implications of the control interface limitations on intent inference are analyzed. The study results show that our approach in many scenarios outperforms existing solutions for intent inference in assistive teleoperation and otherwise performs comparably. Our findings demonstrate the benefit of probabilistic modeling and the incorporation of human agent behavior as goaldirected actions where the adjustable rationality model is user customized. Results further show that the underlying intent inference approach directly affects shared autonomy performance, as do control interface limitations.

The ANEMONE: Theoretical Foundations for UX Evaluation of Action and Intention Recognition in Human-Robot Interaction

“…It is argued that mutual interaction between humans and humans is needed if robots should be considered as partners instead of as tools [ 17 , 19 , 30 , 35 , 36 , 37 ], but to what extent they need to grasp the intentions of others is a much debated issue. However, at least it is argued that to achieve some kind of action and intention recognition between humans and robots, which possibly is a pre-requisite for some basic social interaction skills [ 24 , 26 , 28 , 30 ], is necessary for developing into engaging in more advanced forms of social interaction such as joint actions and mutual collaboration [ 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 ]. In other words, it requires that robots are able to perceive similar emotional and behavioral patterns and environmental cues as humans do (e.g., [ 1 , 5 , 53 , 54 , 55 ]).…”

“…In the fields of HRI and robotics, there has been a lot of research on robots identifying, understanding, and predicting human intention and actions (e.g., [ 1 , 17 , 18 , 19 , 20 , 24 , 33 , 39 , 40 , 41 , 45 , 50 , 51 , 52 ]). It should be noted, however, that there is little existent work where robots are able to fully satisfy the requirements for having recognition capacities, although they display some aspect of recognition.…”

Human-AI Collaboration in Healthcare: A Review and Research Agenda

“…Fourth, there are multiple challenges related to the behavioral and social aspects of human-AI collaboration in healthcare, which also impact their adoption. Other than the lack of trust in AI [50] [31], humans and AI systems are often unable to infer each other's goals and intentions [11], which is necessary to resolve conflicts that may occur. In healthcare, there are also concerns about who is accountable for the outcomes in a human-AI collaboration [21], e.g., a treatment decision that results in a patient's death.…”

A Probabilistic Model for Real-Time Semantic Prediction of Human Motion Intentions from RGBD-Data

“…In contrast to these works, in an indoor environment these cues are typically not available and instead of learning the static context within our work we explicitly model it. Looking at the robotics domain, the idea of considering the plausible alternatives for human intentions is in line with [ 8 , 19 , 25 ]. Within the first of these works, hypotheses about occupied areas in the map are created by considering various trajectories.…”

“…In our work, the discretization of the areas is based on the semantics of the environment. For the second work considering a set of plausible alternatives for human intentions, i.e., [ 25 ], intentions are estimated for assistive robotic teleoperation by hypothesizing about the objects present on a table as potential targets. Contrary to our work, the final destinations are considered as the set of plausible alternatives.…”