Brain-inspired behavioral decision-making of mobile robots based on the motivated developmental network

Area of Science:

• Robotics
• Artificial Intelligence
• Computational Neuroscience

Background:

• Mobile robot navigation in unknown environments presents challenges in computational efficiency and long-term learning.
• Existing methods struggle with high-dimensional spaces and maintaining adaptability over time.

Purpose of the Study:

• To propose an improved behavioral decision-making model for mobile robots.
• To enhance computational efficiency, long-term learning capability, and adaptability in unknown environments.

Main Methods:

• Integration of a deep neural network with a neural remodeling mechanism and an adaptive step-size strategy.
• Weighted integration of motivated developmental network and Deep Q-Network target network for improved Q-value estimation.
• Neural remodeling mechanism to reset neuron age and mitigate learning decline.
• Adaptive step-size strategy for optimized path planning based on proximity to obstacles.

Main Results:

• Enhanced reliability of Q-value estimation and improved training stability and convergence speed.
• Mitigation of learning ability decline, maintaining robot adaptability and learning capacity.
• Optimized path planning performance through dynamic acceleration and deceleration.
• Enhanced flexibility and stability in the overall behavioral decision-making model.

Conclusions:

• The proposed hybrid model significantly improves mobile robot decision-making in unknown environments.
• The integration of deep learning, neural remodeling, and adaptive strategies offers a robust solution for navigation challenges.
• Simulation and physical experiments confirm the model's potential for real-world applications.