This paper presents path planning using learning data-driven controllers and approximate dynamic programming. The obtained optimal data-driven controllers are determined by a lower estimation of the cumulative cost through functional approximators with linear parameterization. The lower estimation is obtained non-iteratively with linear programming by obtaining a bounded and well-conditioned response resulting from the regularization of selected regressors. Additionally, a penalty cost has been assigned for invalid regions (obstacles) or abandonment of the defined workspace. Simulations of the robot with 2D and 3D linear motion were performed to illustrate the generation of object-avoiding trajectories from various starting points to an endpoint in a specific workspace.