Human-Machine Shared Stabilization Control Based on Safe Adaptive Dynamic Programming With Bounded Rationality

This article considers the shared control of bounded rational human behavior with cooperative autonomous machines. For the collaboration of humans and machines, it is crucial to ensure the safety of the interactive process due to the involvement of human beings. First, a barrier-function-based state transformation is developed to ensure full state safety constraints. A level-k$$ k $$ thinking framework is exploited to obtain bounded rationality. Every single level-k$$ k $$ control policy is approximated by using adaptive dynamic programming. Inspired by the theory of human behavior modeling, a probabilistic distribution based on Softmax is utilized to model human behavior, which imitates the uncertainty of human intelligence in the cooperative game. Through the construction of a shared control framework, the control inputs of humans and machines are blended to achieve stabilization safely and efficiently. Finally, simulations are implemented to test the effectiveness of the proposed cooperation architecture. The result demonstrates that full-state asymmetric constraints and stabilization are guaranteed in commonly safety-critical situations, and the shared control framework ensures the safety of the overall system when one of the participants is not safety-aware.