⚡ ErrP BCI — Robotic Arm Grasping

FCz · 256 Hz · LDA Classifier
Robot Scene — 2-DOF Arm
Green-glowing object = TARGET. Robot grasps on command. BCI corrects errors.
Real-time EEG — FCz (Single Trial)
Ne ~80ms · Error Negativity Pe ~300ms · Error Positivity CRN · Correct Negativity
Grand Average EEG (All Trials)
Individual trials fade in background. Averaging cancels noise → ErrP components emerge.
ErrP Classifier & Statistics
Error Detection Confidence (LDA)—
θ=0.5
Awaiting trial…
ErrP BCI loop: User watches robot → robot makes error → brain generates Error-Related Potential at FCz → Ne (negative ~80ms) + Pe (positive ~300ms) → LDA extracts features → correction signal sent if confidence > 0.5.
0
Trials
0
Errors
—
Detect. Acc.
🎲 Auto Mode Each trial the robot randomly grasps the correct or a wrong object (50/50). This is the most realistic simulation — both ErrP and CRN epochs accumulate in the grand average over time.
⚠ Force Error Mode The robot always grasps the wrong object. The brain detects the error every trial, generating clear Ne (~80 ms) + Pe (~300 ms) components in the EEG. Best for observing ErrP morphology in isolation.
✓ Force Correct Mode The robot always grasps the correct object. No error is perceived, so the EEG shows only a small CRN (Correct-Related Negativity). Compare this waveform with Force Error to see how different the two signals are.
💡 Suggested Learning Sequence 1. Run ~5 trials in Force Error → observe Ne + Pe building up in the grand average.
2. Switch to Force Correct → compare the smaller, different-shaped CRN.
3. Switch to Auto → watch the classifier separate the two classes in real time.