In the last decades, visual target tracking has been one of the primary\nresearch interests of the Robotics research community. The recent advances in\nDeep Learning technologies have made the exploitation of visual tracking\napproaches effective and possible in a wide variety of applications, ranging\nfrom automotive to surveillance and human assistance. However, the majority of\nthe existing works focus exclusively on passive visual tracking, i.e., tracking\nelements in sequences of images by assuming that no actions can be taken to\nadapt the camera position to the motion of the tracked entity. On the contrary,\nin this work, we address visual active tracking, in which the tracker has to\nactively search for and track a specified target. Current State-of-the-Art\napproaches use Deep Reinforcement Learning (DRL) techniques to address the\nproblem in an end-to-end manner. However, two main problems arise: i) most of\nthe contributions focus only on discrete action spaces and the ones that\nconsider continuous control do not achieve the same level of performance; and\nii) if not properly tuned, DRL models can be challenging to train, resulting in\na considerably slow learning progress and poor final performance. To address\nthese challenges, we propose a novel DRL-based visual active tracking system\nthat provides continuous action policies. To accelerate training and improve\nthe overall performance, we introduce additional objective functions and a\nHeuristic Trajectory Generator (HTG) to facilitate learning. Through an\nextensive experimentation, we show that our method can reach and surpass other\nState-of-the-Art approaches performances, and demonstrate that, even if trained\nexclusively in simulation, it can successfully perform visual active tracking\neven in real scenarios.\n