Enhancing Human-Robot Interaction with Passive Brain-Computer Interfaces: A Neuroergonomic Approach
Hajiagha Memar, Amirhossein
MetadataShow full item record
An effective and seamless interaction is not just about the accuracy and stability of the robotic system, but rather depends on the capability to dynamically adjust to the task and human cognitive states. The ability to continuously and unobtrusively estimate neuroerganomic aspects of human state during interaction with robots can be used to optimize human-centered interactions. This information can also be used as an additional communication channel to enhance the robot's awareness of human cognition and adapt its behavior accordingly. Passive Brain-Computer Interfaces (BCIs) are promising modalities to establish such implicit communications. Contrary to active BCI systems that aim to explicitly command a system by interpreting the user's intention, passive BCIs implicitly provide the system with the covert aspects of user cognition without the purpose of voluntary control.This thesis investigates the use of passive BCIs in the context of Human-Robot collaboration to accomplish a shared task. In particular, a portable electroencephalogram (EEG) system is used to experimentally explore the application of passive BCIs to two common forms of collaboration. This includes: (i) human-robot teleoperation (remote interaction) and (ii) human-robot cooperative manipulation (physical interaction). In the teleoperation experiment, the main aim is to estimate the operator's performance such as hit-rate, reaction time, and situation awareness based on the operator's physiological signals. In this case, a hybrid-BCI system, combining the EEG with eye-tracking information, is used for data recording while individual differences in visual skills are also considered in the analysis. Moreover, a set of EEG measures from psycho-physiology is used to discriminate between visual and motor task-load with the aim of enhancing adaptation strategy regarding the type of assistance.In the cooperative-manipulation experiment, an admittance controlled robot arm is used to accomplish a set of cooperative manipulation tasks. Since proper admittance parameters are task and subject dependent, EEG patterns are used as an objective measure to estimate the user preferences of physical compliance when interacting with the robot arm. Furthermore, the implication of the proposed cognitive model is illustrated by analyzing the EEG-based estimation of operator cognitive-load when interacting with a variable admittance controller. Finally, a novel approach is presented to quantify motor control difficulty when performing a precise cooperative-manipulation. The results are used to form an EEG-based regression model to identify the level of manipulation difficulty experienced by human operators.Although continuous EEG signals suffer from artifacts and have a low signal-to-noise ratio, our findings indicate that an EEG-based passive BCI is a viable source of information to enhance neuroergonomic aspects of HRI. In general, the proposed EEG-based cognitive models can be utilized in two main directions either as an offline assessment tool for human-centered HRI design or as an online implicit communication channel for robot adaptation. For instance, in the robot-assisted learning of motor skills (e.g., rehabilitation), the learning rate can be maximized by adjusting the robot assistance such that the perceived difficulty remains within a desired range.