NiceBot

NiceBot predicts continuous subjective user ratings of robot behavior by integrating multimodal physiological and robotic data to assess and adapt assistive human-robot interactions.

Key Features:

• Continuous Subjective Rating System: Records continuous user ratings via small thumb movements on a wireless controller alongside physiological signals including EEG, respiration, and ECG.
• Multimodal Data Integration: Captures dry EEG recordings, respiratory patterns, cardiac activity, and robotic joint angles within the Robot Operating System (ROS) for holistic assessment.
• Deep Regression Analysis: Uses deep convolutional neural networks (CNNs) to predict subjective ratings from collected data, with predictions showing higher accuracy when using robotic hand position than EEG, ECG, or respiration signals.
• Adaptation and Learning: Analyzes continuous rating data to enable adaptation of robot behavior in response to user feedback for human-compliant assistive robotic systems.
• Cross-User Model Transferability: Applies transfer learning by adapting pre-trained models to new users with varying prior robot experience, with transfer accuracy improving for users with more experience.

Scientific Applications:

• Assistive robotics evaluation: Provides objective, continuous measures of subjective user perceptions during direct human-robot interactions for assistive system development.
• Human-robot interaction research: Integrates and analyzes diverse data streams to study nuances of human-computer interaction and to improve robotic responsiveness and compliance.

Methodology:

Pilot experiments with three users of different prior robot experience collected continuous ratings and multimodal data (dry EEG, respiration, ECG, robotic joint angles in ROS, and wireless-controller thumb movements); deep convolutional neural network regression models were trained to predict ratings and pre-trained models were adapted to new users via transfer learning.