Developing a robot assisted bed bathing system requires the integration of a diverse range of abilities including perception of the area to be cleaned, planning of a trajectory for wiping across this region of interest, and executing this task in a safe and reliable manner. To accomplish this, we curated a bed bathing dataset using a manikin arm and trained a multimodal perception network capable of leveraging thermal and RGB image data to segment water, soap, and dry skin. Given the results of this segmentation, we demonstrated the ability to clean over the perceived soap, water, and dry regions while ensuring safety through compliance in both the hardware and controls. Limb repositioning is a natural expansion upon this problem as it would allow for washing of hard to reach areas such as under the arm. We can represent this task as an active manipulator moving a passive arm by a fixed grasp point and formulate a dynamics model which can then be used to predict the states of the robot and human arm given a torque command to the robot. In this thesis, this model is verified in simulation (PyBullet) and future work is discussed in the context of assisted bathing.