In the near future, human-robot coexistence and symbiosis will be a common scenario in our society. Especially with the increasing number of patients with stroke or other neurological disorders and the gradually aging population, people may need wearable exoskeletons that actively assist human's movements. In designing these robots, physical humanrobot interaction (pHRI) plays an important role. How to let human and robot cooperatively perform motor tasks and help each other is a grand challenge. Our research establishes a human-robot physical symbiosis framework that biomimics human's behavior when performing interactive motor skills. The human and robot are modeled as two adaptive controllers in parallel with the plant (system under control). As a result, we will have two feedback controllers working together, constantly adapting to each other's behavior and optimally stabilizing the plant to achieve a common goal. In addition, we propose an inverse optimal control method to estimate human control strategy. This information can enable the robot to predict future consensus interactive behaviors in order to cooperate with the human effectively. Experimental verifications have been carried out using double inverted pendulum to simulate a human-robot cooperative balance task in MATLAB environment.