Collaborative Optimization Framework of Interactive Motion Planning and Emergency Protection Control for Human-Robot Interaction

Abstract

Due to the uncertainty of human driving behavior, an autonomous vehicle (AV) may activate its emergency protection policy frequently and unnecessarily when interacting with human drivers, degrading driving efficiency. To address this issue, this paper proposes an optimization framework that coordinates nominal motion planning and emergency protection control, using stochastic model predictive control and a sparse scenario tree to anticipate future collision events. By explicitly considering future safety risks, AV's planner can reduce unnecessary emergency interventions and achieve a balance between safety and performance. The proposed framework integrates backward reachability analysis to precompute the set of unsafe states, and employs recursive Bayesian estimation with a Boltzmann noise-rational human model to predict driver behavior. Simulations demonstrate that the proposed control reduces control costs by 13.1% and improves efficiency by 63.4% compared to baseline methods. Moreover, the interactive experiment platform based on two minicars demonstrates that the proposed method is effective in real-time environments.