A method includes accessing RGB and depth image data representing a scene that includes at least a portion of a robotic limb. Using this data, a computing system may segment the image data to isolate and identify at least a portion of the robotic limb within the scene. The computing system can determine a current pose of the robotic limb within the scene based on the image data, joint data, or a 3D virtual model of the robotic limb. The computing system may then determine a desired goal pose, which may be based on the image data or the 3D virtual model. Based on the determined goal pose, the computing device determines the difference between the current pose and the goal pose of the robotic limb, and using this difference, provides a pose adjustment that for the robotic limb.