SNN-Based Robotic Arm Control Method for Motion Imitation Using EMG and DVS

Spiking Neural Network-Based Robotic Arm Control Technology Combining EMG and Event Cameras

This technology is a neuromorphic control system that receives EMG data from sensors and visual motion information from DVS cameras, converts them into spike signals via adaptive filtering and delta-sigma modulation, and inputs them into a multi-spiking neural network to classify and replicate hand and arm movements in real time.

Conventional control methods based on EMG and acceleration sensors have limitations in precisely mimicking hand and arm movements, suffer from low real-time responsiveness, and consume high power, which hinders the performance advancement of medical robotic systems.

By converting EMG data into spike signals and processing DVS camera data through cropping and down-sampling for parallel input into an SNN model, this technology enhances computational efficiency and enables high-speed, precise motion imitation with low power consumption. It can be applied to prosthetic limbs, rehabilitation robots, and remote-controlled manipulators, offering new possibilities for reducing battery load while instantly reflecting user intent.

‍

‍

Key Features:

• A communication unit that receives EMG data related to human hand or arm movements measured by EMG sensors
• A communication unit configured to also receive information on hand or arm movements detected by a DVS camera
• A component that extracts features from the received EMG signals using an adaptive filter and then performs delta-sigma modulation to convert them into spike signals
• A component that inputs the converted spike signals and the DVS camera's motion information into a trained multi-spiking neural network model

‍

This invention was developed with support from the Ministry of Science and ICT for research and development on neuro-chip design technology and neuro-computing platforms that mimic the human nervous system.