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

Behavioral decision-making for mobile robots in unknown environments is a key research topic in the field of robotics applications. To address issues such as low computational efficiency in high-dimensional spaces and diminished long-term learning capability, this paper proposes an improved method that integrates deep neural network, a neural remodeling mechanism, and an adaptive step-size strategy. First, the motivated developmental network is fused with the target network of the Deep Q-Network through weighted integration, enhancing the reliability of Q-value estimation and improving training stability and convergence speed. Second, a neural remodeling mechanism is designed to reset neuron age, thereby mitigating the decline in learning ability caused by excessive neuron activations and enabling the robot to maintain strong adaptability and learning capacity over time. Furthermore, an adaptive step-size strategy is designed, allowing the robot to accelerate when far from obstacles and decelerate for precise avoidance when nearby, thus optimizing path planning performance. Finally, the integration method of the hybrid model is improved by combining the similarity metric with trajectory evaluation indicator to dynamically modulate the outputs of the cerebellum and basal ganglia, thereby enhancing the flexibility and stability of overall behavioral decision-making model. Simulation and physical experiments validate the potential of the proposed model.