The performance of coherent symbol detection in multiple-input multiple-output (MIMO) systems is tightly related to the channel knowledge at the receiver side. For MIMO detection, achieving near-optimal performance with a low computational complexity has been a topic of several recent researches. In this paper, we propose a robust detector for uncoded MIMO systems with imperfect CSI (ICSI). The proposed detector is based on the likelihood ascent search (LAS) algorithm which achieves near maximum likelihood (ML) performance with a reduced complexity. To take into account the channel estimation error, we propose to use a modified likelihood metric function. Moreover, to enhance the system performance, we suggest to perform the channel estimation and the data detection in an iterative manner. Simulation results show that the proposed scheme outperforms the conventional LAS detector.