Robot-assisted ultrasound (US) offers several technological advances over manual US scanning that can more effectively assist in clinical diagnosis and procedure guidance. However, the adoption of robotic US devices has been hindered by issues of portability and automation, especially in followed-up diagnosis. In this article, we develop a portable robotic system with built-in intelligence for autonomous US acquisition in follow-up diagnosis. Different from the widely used serial manipulator-based US device, this device is built using a 6-RSU parallel mechanism with compact size and maneuverability. We incorporate a force control algorithm and the robot-specific pose feedback mechanism to ensure safety and maintain image quality. Exploring possibilities for intelligence, we propose a follow-up diagnosis workflow tailored to specific clinical applications and present a multimodal reinforcement learning (RL) algorithm for automatic scanning based on a novel similarity network as reward. Our proposed methods is suitable for any target images instead of a few specific ones which increasing autonomous US acquisitions ability for US device. Our simulation experiments show a remarkable success rate of 98.21%, and phantom experiments validate the effectiveness and robustness of the robotic device system with a structural similarity index (SSIM) of 0.764 and a normalized cross correlation (NCC) of 0.977. To the best of our knowledge, our robotic US device is the first device with a parallel mechanism and force-position perception for follow-up diagnosis, featuring automatic acquisition of multiple target US images