Motion planning algorithms for dynamic environments that explicitly take into account actuator constraints require a lot of computational effort due to replanning or optimizing trajectories. This makes them limited in use, especially for autonomous reactive behaviors that need to be computed on-board. Motivated by this need, we present a new real-time method for reactive collision avoidance for systems with bounded curvature in static and dynamic environments. Our approach relies on the implementation of a local steering law that satisfies a predefined bound on path curvature. The steering law depends on a user-defined parametric function that determines the transition between obstacle-free motion and collision avoidance by enforcing an obstacle impenetrability constraint. As such, we propose a systematic procedure which modulates the velocity vector to enforce curvature constraints in complex 2D environments characterized by static and moving obstacles. We provide theoretical guarantees for collision avoidance and we demonstrate this methodology through simulations.