Neural network output feedback control of robot formations.

Area of Science:

• Robotics
• Control Systems
• Artificial Intelligence

Background:

• Existing kinematic-based formation controllers often simplify robot dynamics.
• Accurate estimation of robot velocities and dynamics is crucial for stable formations.
• Collision avoidance in multi-robot systems remains a significant challenge.

Purpose of the Study:

• To develop a robust leader-follower formation control law that accounts for robot dynamics.
• To enhance formation stability and collision avoidance using advanced control techniques.
• To validate the proposed control strategy through numerical simulations.

Main Methods:

• A combined kinematic/torque output feedback control law based on backstepping.
• Integration of a neural network (NN) for approximating robot dynamics with online weight tuning.
• Design of a novel NN observer for estimating linear and angular velocities.
• Lyapunov stability theory to analyze formation error bounds and stability.

Main Results:

• The proposed control law ensures uniformly ultimately bounded errors for the entire formation, relaxing the separation principle.
• The NN observer effectively estimates follower and leader robot velocities.
• Stability analysis confirms formation robustness in the presence of obstacles.
• Collision prevention within the formation is achieved by treating robots as obstacles.

Conclusions:

• The developed control law offers a more accurate and stable approach to leader-follower robot formation control compared to purely kinematic methods.
• The use of neural networks and observers enhances the system's ability to handle complex dynamics and uncertainties.
• The strategy effectively prevents intra-formation collisions and maintains formation stability around obstacles.