We experimentally generate polarization spectroscopy (error) signals corresponding to the D2-line hyperfine transitions, F-g = 2 -> F-e = 1, 2, 3 of Rb-87, and F-g = 3 -> F-e= 2, 3, 4 of Rb-85, and show that the strongest error signals correspond to the closed hyperfine transitions (oscillator strength similar to 0.7), F-g = 2 -> F-e = 3 and F-g = 3 -> F-e = 4, respectively. We make the generated error signal robust to fluctuations in external parameters by finding optimum values for the vapor cell temperature and pump-probe intensities at two different beam diameters. We further employ these optimized error signals to directly (without the need for frequency/phase modulation) lock the laser frequency-reducing the rms drift/linewidth from similar to 10 MHz to <500 kHz, when measured over a similar to 60 min duration. In addition, by comparing theoretically calculated and experimentally measured error signals for the pump intensities ranging from similar to 0.1I(sat0)-10(sat0) (where, I-sat0 similar to 1.6 mW cm(-2) is the two-level saturation intensity for the Rb D-2-line transitions), we discuss the applicability and limitations of existing numerical and analytical approaches based on a full multi-level rate equation model for polarization spectroscopy.