Perceived-Danger-Aware Adaptive Robot Behavior: A Framework for Real-Time Human-State-Responsive Interaction
Project Description
Robots operating in close proximity to people need to account not only for physical safety, but also for how their behavior is perceived by humans. Previous studies investigate how different sources of perceived danger shape human physiological and subjective responses.
Building on these findings, this project aims to develop a closed-loop framework for human-aware robot adaptation. The framework will continuously estimate a person’s real-time state of perceived danger from multimodal behavioral and physiological signals, and use this estimated state to adapt the robot’s behavior proactively.
The framework will establish a pipeline from perceived danger → human-state estimation → robot adaptation. Multimodal signals, such as heart rate, heart-rate variability, galvanic skin response, pupil responses, and gaze behavior, will be used to estimate changes in the user’s state during robot interaction. Based on the estimated level and potential source of perceived danger, the robot will dynamically adjust relevant behavioral parameters, such as motion speed, distance, trajectory, predictability, or explanatory feedback.
The goal is to move beyond robots that respond only to explicit user feedback or predefined safety rules toward robots that can infer emerging human discomfort in real time and adapt their behavior accordingly. The resulting framework will provide a generalizable approach for designing robots that are not only physically safe, but also perceptually safe, responsive, and trustworthy during human–robot interaction.
Building on these findings, this project aims to develop a closed-loop framework for human-aware robot adaptation. The framework will continuously estimate a person’s real-time state of perceived danger from multimodal behavioral and physiological signals, and use this estimated state to adapt the robot’s behavior proactively.
The framework will establish a pipeline from perceived danger → human-state estimation → robot adaptation. Multimodal signals, such as heart rate, heart-rate variability, galvanic skin response, pupil responses, and gaze behavior, will be used to estimate changes in the user’s state during robot interaction. Based on the estimated level and potential source of perceived danger, the robot will dynamically adjust relevant behavioral parameters, such as motion speed, distance, trajectory, predictability, or explanatory feedback.
The goal is to move beyond robots that respond only to explicit user feedback or predefined safety rules toward robots that can infer emerging human discomfort in real time and adapt their behavior accordingly. The resulting framework will provide a generalizable approach for designing robots that are not only physically safe, but also perceptually safe, responsive, and trustworthy during human–robot interaction.
Supervisor
SHAO, Qijia
Quota
2
Course type
UROP1000
UROP1100
UROP2100
UROP3100
UROP3200
UROP4100
Applicant's Roles
The applicant will be actively involved in the design, implementation, and evaluation of a human-aware adaptive robot interaction framework. Primary responsibilities include:
● Processing multimodal human-state data, including physiological and behavioral signals such as heart rate, heart-rate variability, galvanic skin response, pupil responses, and gaze behavior.
● Supporting the development and evaluation of machine learning or statistical models for real-time human-state and perceived-danger estimation.
● Assisting in the design of adaptive robot behaviors, such as adjusting motion speed, interpersonal distance, trajectory, predictability, or explanatory feedback based on the estimated human state.
● Implementing and testing the closed-loop interaction pipeline from human-state estimation to adaptive robot behavior.
● Conducting experimental evaluations to assess whether adaptive behaviors can reduce perceived danger and improve the safety and trustworthiness of human–robot interaction.
● Contributing to data analysis, visualization, documentation, and potential paper writing and submission.
● Processing multimodal human-state data, including physiological and behavioral signals such as heart rate, heart-rate variability, galvanic skin response, pupil responses, and gaze behavior.
● Supporting the development and evaluation of machine learning or statistical models for real-time human-state and perceived-danger estimation.
● Assisting in the design of adaptive robot behaviors, such as adjusting motion speed, interpersonal distance, trajectory, predictability, or explanatory feedback based on the estimated human state.
● Implementing and testing the closed-loop interaction pipeline from human-state estimation to adaptive robot behavior.
● Conducting experimental evaluations to assess whether adaptive behaviors can reduce perceived danger and improve the safety and trustworthiness of human–robot interaction.
● Contributing to data analysis, visualization, documentation, and potential paper writing and submission.
Applicant's Learning Objectives
By participating in this project, the applicant will:
● Gain hands-on experience in human–robot interaction (HRI) research and human-centered robotics.
● Develop practical skills in processing and analyzing multimodal physiological and behavioral data.
● Learn how physiological and behavioral signals can be used to infer human cognitive and affective states during interaction.
● Gain experience with machine learning and statistical methods for real-time human-state estimation.
● Gain experience in evaluating adaptive robotic systems through user studies and quantitative analysis.
● Improve research, communication, and collaboration skills through participation in a multidisciplinary HRI research environment.
● Gain hands-on experience in human–robot interaction (HRI) research and human-centered robotics.
● Develop practical skills in processing and analyzing multimodal physiological and behavioral data.
● Learn how physiological and behavioral signals can be used to infer human cognitive and affective states during interaction.
● Gain experience with machine learning and statistical methods for real-time human-state estimation.
● Gain experience in evaluating adaptive robotic systems through user studies and quantitative analysis.
● Improve research, communication, and collaboration skills through participation in a multidisciplinary HRI research environment.
Complexity of the project
Moderate