The idea of extending and augmenting the capabilities of the human body has been an enduring area of exploration in fiction, research, and industry alike. The most concrete realizations of this idea have been in the form of wearable devices such as prostheses and exoskeletons, that replace or enhance existing human functions. With recent advances in sensing, actuation, and materials technology, we are witnessing the advent of a new class of wearable robots: Supernumerary Robotic (SR) devices that provide additional degrees of freedom to a user, typically in the form of extra limbs or fingers. The development, analysis, and experimental evaluation of one such SR device, a Wearable Robotic Forearm (WRF) for close-range collaborative tasks, forms the focus of this dissertation. We initiated its design process through a basic prototype mounted on a user's elbow, and conducted an online survey, a contextual inquiry at a construction site, and an in-person usability study to identify usage contexts and functions for such a device, and formed guidelines for improving the design. In the next WRF prototype, we added two more degrees of freedom while remaining within acceptable human ergonomic load limits, and expanding its reachable workspace volume. We then developed the final prototype based on further feedback from a pilot interaction study, and found an analytical solution for its inverse kinematics. Going beyond static analyses with predefined robot trajectories, we further addressed the biomechanical effects of wearing the WRF using a detailed musculoskeletal model, and developed a motion planner that minimizes loads on the user's muscles. Looking at the other side of the physical interaction between the user and WRF, we applied human motion prediction and feedback control for stabilizing the robot's end- effector position when subjected to disturbances from the wearer's body movements. Finally, we conducted a user study involving a collaborative pick-and-place task with the WRF acting in two conditions: responding to direct speech commands from the wearer, and predicting human intent using supervised learning models. We evaluated the quality of interaction in the two conditions through human-robot fluency metrics. The WRF, and its associated systems described in this dissertation do have limitations, particularly in terms of ergonomics, feedback control performance, and fluency of interaction. However, as a prototype, the WRF shows that SR devices can be effective agents in human-robot collaboration when they possess capabilities for mutual adaptation while reducing the cognitive load on the user.