The accurate localization of wireless devices plays an important role in several real-time Internet of Things (IoT) applications. In a network composed of many IoT sensors, a distributed collaborative localization approach can give more accurate localization performance based on a decentralized and low-complexity processing. However, the presence of Non-Line of Sight links between IoT devices detrimentally impacts the localization accuracy. In this paper, we propose a distributed localization algorithm based on a convex relaxation of the Huber loss function. Moreover, to reduce the algorithm convergence time, an iterative stochastic gradient descent algorithm is proposed. Through numerical simulations, we show that the proposed algorithm when used with optimal relaxation parameters of the Huber loss function achieves very low root mean square error and outperforms existing algorithms in the literature. Finally, we validate our proposed scheme using real experimental data.