Shrinkable corridor: A collision-free obstacle constraint method for optimization-based trajectory planning

Abstract

Optimization-based trajectory planning is widely used in autonomous driving, yet efficient obstacle avoidance for vehicles with non-negligible dimensions remains challenging. This paper proposes a shrinkable-corridor method within a MINCO-based trajectory optimization framework. The free space is represented by overlapping convex polygons, which are adaptively shrunk according to the vehicle approximation model. In open regions, a two-disc model enables simple center-point constraints, while in narrow regions, a compact approximation with front-back separated constraints improves feasibility. To handle possible collisions between finite constraint points, an iterative refinement strategy detects risky segments, inserts local constraints, and updates the corresponding corridors. The proposed method reduces constraint complexity while enforcing full-dimensional collision constraints at sampled points and empirically removing inter-sample collisions through iterative checking. Simulations demonstrate improved optimization efficiency and trajectory feasibility compared with representative safe-corridor baselines.