Control Profile Derivation for Collision Imminent Steering via Solutions of Large-Scale Optimization Problem

In collision-imminent situations where braking alone is insufficient, integrated steering control is an effective avoidance measure. However, a generalized strategy that minimizes the required avoidance distance by fully leveraging a vehicle’s physical limits has not been established, as prior studies have focused on computational challenges or proposing new architectures. This paper proposes a framework to derive such a strategy by defining a large-scale optimization problem that integrates steering, traction, and braking. The most effective control profiles are identified from numerous solutions using a clustering algorithm. This data-driven approach overcomes the initial value dependency of non-convex optimization, enabling the discovery of globally optimal strategies. The derived optimal strategy exhibits novel control methods including the use of traction force near an obstacle and enables collision avoidance at shorter distances than conventional real-time controllers. These results suggest that a control structure with separated path generation and tracking is effective for implementing the proposed strategy. This study provides a foundational reference for designing advanced controllers for collision-imminent situations and a base framework for various derivative studies.

Hali tarjima qilinmagan