Bio-Inspired Proprioception for Sensorless Control of a Klann Linkage Robot Using Attention-LSTM.

Area of Science:

• Robotics
• Artificial Intelligence
• Biomimetics

Background:

• Walking robots face challenges in precise gait control due to reliance on complex sensors and control systems.
• Commercialization and lightweight designs are hindered by current sensor and control architecture limitations.

Purpose of the Study:

• To develop an AI-based, sensorless feedback control framework for walking robots.
• To incorporate biological proprioception principles into robotic control.
• To enable adaptable sensing for complex terrains.

Main Methods:

• Developed a walking robot utilizing the Klann linkage's morphological intelligence.
• Created a dataset using motor current signals as 'interoceptive sensing' and synchronized with angular data.
• Trained an Attention-LSTM (A-LSTM) model to predict motor states from internal current data.

Main Results:

• The A-LSTM model successfully decoded nonlinear physical information from internal current data.
• A stable biomimetic walking loop was achieved by integrating the A-LSTM model into a PI controller.
• The system demonstrated effective sensorless feedback control, eliminating the need for external position sensors.

Conclusions:

• The proposed AI-based sensorless framework successfully mimics biological proprioception for robotic control.
• This approach overcomes limitations of traditional sensor-dependent systems, paving the way for lighter and more adaptable walking robots.
• The study establishes a novel sensing paradigm for robots operating in complex environments.