A method for manipulating a multi-link robotic arm includes: at a first time, recording a first optical image through an optical sensor arranged proximal a distal end of the robotic arm proximal an end effector; detecting a global reference feature in a first position in the first optical image; virtually locating a global reference frame based on the first position of the global reference feature in the first optical image; calculating a first pose of the end effector within the global reference frame at approximately the first time based on the first position of the global reference feature in the first optical image; and driving a set of actuators within the robotic arm to move the end effector from the first pose toward an object keypoint, the object keypoint defined within the global reference frame and representing an estimated location of a target object within range of the end effector.