Emotion inspired cognitive architecture for robotic adaptive path planning
Zheming Zhang, Will Neil Browne, Dale A. Carnegie · QUT ePrints (Queensland University of Technology) · 2018
Adaptable navigation is critical to extend the range of applications for mobile robots in daily life. An ideal architecture of mobile robots should adapt its path-planning methods to various navigation scenarios. But traditional architectures are static, plus they often need hand-coded prior knowledge (e.g. tuned hyperparameters for targeted scenarios) to ensure the path-planning methods function well. This paper proposes an adaptive architecture through a hyperparameter-adjustment approach for robotic path-planning tasks. The architecture can automatically learn adaptive path-planning knowledge, termed Planning-Action-Outcome Contingency (PAOC), which is inspired by emotion theories in cognitive neuroscience. PAOC knowledge is learned in a pattern of human interrogable "if-then" rules by an Accuracy-based Learning Classifier System (XCS) algorithm. Navigation simulations of a mobile robot were conducted within 31 differing scenarios. Results show the proposed architecture achieved adaptive path-planning by automatically learning PAOC patterns in all the scenarios. These PAOC patterns also provide visual interpretations regarding what the robot perceives in the scenarios.