EEG-based Action Anticipation in Human-Robot Interaction
Author Information
Author(s): Rodrigo Vieira, Plinio Moreno, Athanasios Vourvopoulos
Primary Institution: Instituto Superior Técnico, Lisbon, Portugal
Hypothesis
Can EEG signals be used to anticipate human actions in human-robot interactions?
Conclusion
The study demonstrates that EEG combined with deep learning can predict human actions several hundred milliseconds before execution, enhancing human-robot collaboration.
Supporting Evidence
- EEG signals can detect brain activity preceding movement by over a second.
- Deep learning models achieved up to 80.90% accuracy in predicting actions.
- The MLP classifier provided a decision time advantage of around 590 milliseconds.
Takeaway
This study shows that robots can learn to guess what a person will do next by reading their brain signals, helping them work better together.
Methodology
The study evaluated multiple deep learning models on EEG data to predict human actions during motor imagery and execution tasks.
Potential Biases
Potential biases due to the small number of participants and the specific conditions of the experiments.
Limitations
The study had a small sample size and the pilot experiment may have introduced movement artifacts that affected EEG signal quality.
Participant Demographics
12 healthy, BCI-naïve subjects participated in the study.
Statistical Information
P-Value
p = 0.012 for MLP vs CNN_shallow; p < 0.001 for MLP vs CNN_LSTM
Confidence Interval
95% CI: 520–870 ms for MLP DTA
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website