A Brain-Inspired Approach for Probabilistic Estimation and Efficient Planning in Precision Physical Interaction.

Area of Science:

• Neuroscience
• Robotics
• Artificial Intelligence

Background:

• Simulating complex physical interactions requires integrating multiple cognitive functions.
• Existing models often struggle with real-time prediction and adaptive control.

Purpose of the Study:

• To propose a novel spiking neural network (SNN) architecture that mimics the joint function of multiple brain regions for precision physical interactions.
• To enhance movement planning, contact prediction, and radial compensation capabilities in artificial systems.

Main Methods:

• A double recurrent network inspired by the hippocampus for spatiotemporal contact prediction.
• A cerebellum-inspired module for temporally dynamic radial contact response.
• A prefrontal cortex-like block-based feedforward network for movement planning.

Main Results:

• The integrated SNN successfully simulates the joint cognitive functions of prediction, control, and planning.
• The proposed controller and planner generate effective teaching signals for reinforcement learning.
• Experimental validation confirms the efficacy of the novel SNN structure.

Conclusions:

• The developed SNN architecture provides a biologically plausible and effective model for complex physical interactions.
• This approach advances the integration of cognitive neuroscience principles into robotics and AI.
• The method offers a feasible network initialization strategy for reinforcement learning in physical tasks.