Bio-Inspired Tiny AI for Mini Robots
Abstract (300 words maximum)
Recently, robotic scientists have miniaturized robots into millimeter level sizes by designing new actuators with MEMS technology and novel materials. However, most of these robots can merely receive control signals passively and actuate. Their intelligence is inferior to natural insects of the same size. On the other hand, recent advances in tiny ML (Machine Learning) have brought compact machine learning models to ultra-low-power and resource-limited devices, such as general-purpose micro-controller units. However, these works primarily target visual or audio recognition tasks instead of sensing, decision-making, or acuation control for small robots. The goal of our work is to achieve AI functions of small robots with limited power, memories, and computing resource of MCU or microprocessor. Specifically, we explore the brain inspired computing algorithm, spiking neural network, that can smoothly operate on an MCU and adapt gait patterns for the locomotion of quadruped robots. In the future, our work aims at the end-to-end AI application for robots, namely for sensing actuation.
Academic department under which the project should be listed
SPCEET - Robotics and Mechatronics Engineering
Primary Investigator (PI) Name
Yan Fang
Bio-Inspired Tiny AI for Mini Robots
Recently, robotic scientists have miniaturized robots into millimeter level sizes by designing new actuators with MEMS technology and novel materials. However, most of these robots can merely receive control signals passively and actuate. Their intelligence is inferior to natural insects of the same size. On the other hand, recent advances in tiny ML (Machine Learning) have brought compact machine learning models to ultra-low-power and resource-limited devices, such as general-purpose micro-controller units. However, these works primarily target visual or audio recognition tasks instead of sensing, decision-making, or acuation control for small robots. The goal of our work is to achieve AI functions of small robots with limited power, memories, and computing resource of MCU or microprocessor. Specifically, we explore the brain inspired computing algorithm, spiking neural network, that can smoothly operate on an MCU and adapt gait patterns for the locomotion of quadruped robots. In the future, our work aims at the end-to-end AI application for robots, namely for sensing actuation.