Error-related potentials-based human-robot intelligent system
- Publication Type:
- Thesis
- Issue Date:
- 2022
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Brain-Computer Interface (BCI) is an emerging technology that provides natural and direct communication between humans and machines. Recent BCI works aimed to create accurate and reliable BCI systems in the field of Human-Robot Interaction (HRI). Of these, the BCI paradigm based on error-related potentials (ErrPs), a cognitive phenomenon derived from EEG signals, is particularly promising. ErrPs are involuntarily evoked when a person perceives unexpected errors in the environment. Unlike other BCI paradigms that require users to actively imagine the mental commands or engage with additional visual stimuli, ErrPs depends on the user’s experience on assessing the correctness of the robot behaviours. The ErrP-based BCI does not require additional training and does not interrupt the user’s original workflow. This thesis presents two novel ErrP-based BCI systems:
First, a novel robotic design for ErrP-based BCI that allows humans to evaluate the robot’s intentions continuously. Current ErrP-based BCI cannot handle interaction sequences that involve continuous robot movements. For example, it is difficult to extract a time-locked event when the user detects an unexpected error while the robot arm is already in motion. The high classification accuracy (77.57%) from the first system confirmed that the proposed ErrP-based BCI paradigm allows continuous evaluation of robot intentions in real-time and thus enable earlier intervention before the robot commits an error.
Second, an ErrP-based shared autonomy via deep recurrent reinforcement learning is developed. Current BCI systems use ErrP as either an implicit control signal to the agent or a reward signal in reinforcement learning (RL). The novel proposed framework using ErrP as an input feature in the trained RL model enables human intervention with a trained autonomous agent. In a simulation with 70% ErrP accuracy, agents completed the task 14.1 % faster. In the real-world experiment, agents completed the navigation task 14.9% faster. The evaluation results confirmed that the shared autonomy via deep recurrent reinforcement learning is an effective way to deal with uncertain human feedback in a complex HRI task.
These two novel BCI systems advance the current ErrP-based BCI capabilities and enable a wide range of new interaction possibilities between human and robot. This thesis represents an important step toward a BCI-based shared autonomy between humans and robots.
Please use this identifier to cite or link to this item: