Massive multiple-input multiple-output (MIMO) is a core technology for 5G and beyond systems. However, symbol detection in massive MIMO requires high complexity matrix inversions. To tackle this problem, a novel and robust low complexity hybrid algorithm (HA) is proposed for uplink symbol detection in massive MIMO systems with a large number of users. Proposed HA integrates two novel techniques; non-stationary Newton iteration (NSNI) and improved sequential Richardson iteration (ISRI), which are proposed in this paper. Newton iteration (NI) is a promising technique for approximate matrix inversion, however, in this paper, Newton iteration (NI) is realized as the stationary iterative method which uses constant step size for all iterations. Consequently, NI suffers from performance-complexity trade-off. To address this issue, NSNI is proposed, which utilizes non-stationary step size that changes at each iteration. Moreover, Richardson iteration is a simple but efficient algorithm for massive MIMO detection, however, RI suffers from intersymbol interference (ISI) which is a major reason for the low performance of RI when the number of users scales up in massive MIMO system. Hence, symbols are updated sequentially to extenuate ISI in RI. In addition, to further improve the performance of RI, optimal step sizes based on each symbol-index in RI are computed and hence, an improved stationary Richardson iteration (ISRI) is introduced. Finally, to further boost bit error rate (BER), NSNI and RI are integrated into pseudo-stationary iterative HA for low complexity symbol detection in massive MIMO systems. Simulation results validate low complexity, superior BER performance and robustness of proposed HA as compared to recently reported several massive MIMO detection techniques, under both perfect and imperfect channel state information at the receiver.