T1ρ-mapping using mono- or bi-exponential models usually require multiple spin-lock times (#TSLs). Choosing the optimal #TSLs for improved SNR, i.e. minimizing the Cramer-Rao lower bound (CRLB), is important. However, Gaussian statistics with CRLB usually lead to the repetition of longer TSLs and the minimum number of shorter TSLs. This choice is non-robust to large data acquisition errors caused by subject motion or other scan related problems that strongly affect quantitative parameters. To alleviate this problem, we propose a robust T1ρ-protocol based on optimized #TSLs using CRLB with robust statistics, and outlier-robust fitting method.