Optimal Robot-Environment Interaction Under Broad Fuzzy Neural Adaptive Control.

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Conventional fuzzy neural networks (NN) struggle with determining adequate NN units for unknown dynamic models.
• Robotic systems require robust control strategies for safe and effective interaction with uncertain environments.
• Ensuring system stability and performance under state constraints is a critical challenge in advanced robotics.

Purpose of the Study:

• To propose a novel control strategy using a broad fuzzy neural network (BFNN) for robots interacting with unknown environments.
• To develop an adaptive impedance learning scheme for optimal robot-environment interaction without prior environment modeling.
• To address state constraints in trajectory tracking using a barrier Lyapunov function (BLF).

Main Methods:

• Utilizing the broad learning system (BLS) to overcome NN unit selection challenges in BFNN.
• Implementing adaptive impedance learning for compliant and optimal robot-environment interaction.
• Integrating a barrier Lyapunov function (BLF) into the trajectory tracking controller to handle state constraints.

Main Results:

• The proposed BFNN strategy effectively approximates unknown dynamic models.
• Adaptive impedance learning facilitates a soft, compliant contact scheme.
• The controller ensures desired tracking and interaction performance while maintaining closed-loop system stability.
• Simulations and experiments on a 2-DOF manipulator and Baxter robot validate the BFNN's effectiveness.

Conclusions:

• The BFNN-based control strategy offers a robust solution for robotic interaction in unknown environments.
• The integration of BLS and BLF enhances control performance and system stability.
• The developed method provides a model-free approach for adaptive impedance control in robotics.