BrainBotics

BrainBotics was inspired by a simple but powerful problem: millions of people living with paralysis, ALS, spinal cord injuries, or severe motor disabilities often depend on others for even basic daily tasks. Existing assistive technologies are either extremely expensive, invasive, or physically demanding. We wanted to explore whether human intention alone—captured through brain signals—could be translated into real-world actions. This led us to build a brain-computer interface that decodes motor intention directly from EEG signals. BrainBotics is a non-invasive Brain-Computer Interface (BCI) proof-of-concept that converts EEG brain signals into machine-understandable commands. The system takes raw motor imagery EEG data, processes neural activity from the motor cortex, extracts meaningful features, and uses deep learning to classify user intention such as left-hand or right-hand movement imagination. These predictions are then translated into robotic control commands and visualized through a live dashboard. BrainBotics is a non-invasive Brain-Computer Interface (BCI) proof-of-concept that converts EEG brain signals into machine-understandable commands. The system takes raw motor imagery EEG data, processes neural activity from the motor cortex, extracts meaningful features, and uses deep learning to classify user intention such as left-hand or right-hand movement imagination. These predictions are then translated into robotic control commands and visualized through a live dashboard. Working with raw EEG data presented multiple challenges. EEG signals are highly noisy, subject-dependent, and sensitive to artifacts such as muscle movement or eye blinks. We faced issues with inconsistent epoch lengths across subjects, data preprocessing failures, class imbalance, and training time constraints during the hackathon. Building a real-time dashboard while maintaining inference speed and visualization stability also required significant optimization. Working with raw EEG data presented multiple challenges. EEG signals are highly noisy, subject-dependent, and sensitive to artifacts such as muscle movement or eye blinks. We faced issues with inconsistent epoch lengths across subjects, data preprocessing failures, class imbalance, and training time constraints during the hackathon. Building a real-time dashboard while maintaining inference speed and visualization stability also required significant optimization. We successfully built an end-to-end BCI pipeline capable of decoding human motor intention from raw EEG data. Our system processed data from 20 subjects and over 1700+ epochs, achieving our best results using EEGNet with approximately 0.63 F1 score. We also developed a live visualization dashboard that demonstrates real-time EEG activity, confidence scores, latency metrics, and command generation. Most importantly, we proved that a thought-guided assistive system is technically feasible even without live EEG hardware. This project taught us that building a Brain-Computer Interface requires the intersection of neuroscience, signal processing, machine learning, and real-time systems engineering. We learned how motor cortex rhythms such as Mu and Beta bands represent movement intention, how preprocessing dramatically affects model performance, and why deep learning architectures like EEGNet are better suited for EEG classification than many traditional approaches.