This paper presents the implementation of a soccer mobile robot agent that uses vision based object detection procedures. The soccer mobile robot is based on a Google Android smartphone and takes advantage of its computation and sensing capabilities. The agent implementation uses a multi-threading methodology to generate different agent processes such as the vision sense, the decision-making, and the application manager. The vision sense thread captures images from the smartphone camera and provides environment information using Renderscript computation while the decision-making thread provides effective soccer agent behavior. This paper presents and discusses the agent implementation and the effectiveness of the vision based object detection procedures.