Abstract: Estimating individualized Treatment Rules (ITRs) is fundamental in causal inference, particularly for precision medicine applications. Traditional ITR estimation methods rely on inverse probability weighting (IPW) to address confounding factors and $L_{1}$-penalization for simplicity and interpretability. However, IPW can introduce statistical bias without precise propensity score modeling, while $L_{1}$-penalization gives computational bias and requires subgradient methods, which slows down the convergence of optimization algorithms for ITR estimation. In this paper, we propose a novel ITR estimation framework formulated as a weighted convex optimization problem. The weights are obtained via model-free distributional covariate balancing, and instead of soft $L_{1}$-penalization, we use hard $L_{1}$-ball constraint, allowing the objective to maintain smoothness. The optimal ITR can be robustly and effectively computed by projected gradient descent (PGD). We provide a comprehensive analysis of the proposed framework including global convergence and complexity guarantee of the PGD algorithm and statistical estimation guarantee under natural generative ITR models. Extensive simulations and applications demonstrate that our framework achieves significant gains in both robustness and effectiveness for ITR learning against several existing methods.