Dynamic Sequence Encoding Mechanism Integrated Drivable Domain Trajectory Optimization for Embodied Intelligent Driving

Embodied intelligent driving systems in dynamic scenarios still face challenges in representing driving-sequence switching, insufficient coupling between the drivable domain and driving tasks, and limited continuity of trajectory planning. To address these issues, this paper proposes a drivable domain trajectory optimization method integrated with a dynamic sequence encoding mechanism. Typical driving tasks, including cruising, car-following, and overtaking, are uniformly modeled as structured driving sequences comprising basic constraints, sequence boundaries, target evolution, cost weights, and switching conditions, enabling driving tasks to serve as unified planning inputs for subsequent drivable domain construction and trajectory optimization. A sequence-driven task-oriented drivable domain reshaping mechanism is constructed to dynamically adjust the lateral path boundaries and longitudinal advance boundaries according to the currently activated sequence, thereby reshaping the fundamental drivable domain into an executable region matched to the current driving sequence. Furthermore, a sequence encoding and decoding mechanism is established to realize a structured mapping between dynamic traffic states and driving sequences, while uniformly representing the evolution of objectives, weights, and constraint conditions during sequence switching. To suppress frequent switching that may occur during multi-sequence transitions, hysteresis and minimum dwell time constraints are introduced, and multiple Lyapunov functions are employed to derive a sufficient condition for switching stability. Simulation results show that, in the two tested simulation scenarios, the proposed method achieves continuous switching among cruising, car-following, and overtaking sequences while maintaining favorable longitudinal and lateral motion smoothness. In the tested three-lane dynamic-obstacle scenario, compared with Algorithm 1, the peak deceleration amplitude, peak steering angular acceleration, and peak front-wheel steering angle are reduced by approximately 62.2%, 63.3%, and 46.9%, respectively; compared with Algorithm 2, the corresponding reductions are approximately 37.0%, 14.4%, and 67.6%, respectively. The results indicate that, under the tested scenarios and comparison settings, the proposed method can effectively coordinate driving-sequence switching with local trajectory planning while improving the smoothness and stability of vehicle motion.