Neuromuscular and sensorimotor degeneration significantly reduce physical, cognitive, and social well-being across the life span. Assistive technologies have proven useful in augmentation or supporting patients as they work towards positive therapeutic outcomes. However, the efficacy of these tools is undermined by a lack of quantifiable feedback regarding the degree of intervention at one point and the effectiveness of various training parameters (such as timing, intensity, etc.). In this dissertation, the human-robot interaction (HRI) is studied to establish an efficient, safe, and quantifiable therapeutic approach that accounts for variation in biomechanical and bio-signal parameters. The research involves five main contributions: (1) identifying effective task-based trajectories for rehabilitation while performing different tasks within the range of the targeted limb, (2) developing an intervention prediction package based on human factors, (3) establishing an interfacing system to reproduce the selected effective trajectories through robots with an assist-asneeded (AAN) intervention protocols using the predictive model, (4) any possible complication due to using such interventions either through robot end effector, or exoskeletons is assessed from the human anatomy prospective; and (5) the same AAN protocol is implemented in a human robot collaboration task where muscle fatigue affects performance of the human counterpart and accordingly robot is required to adjust its contribution. Preliminary results have indicated that both the task and the interaction forces are crucial in recruiting and affecting the muscles; A predictive model based on the interaction force and set EMG threshold values is derived through experiments in the cases of no assistance, partial assistance, and full assistance scenarios and is employed to drive contributing factor ( ) to modify the Euler-Lagrange equation and implement the AAN protocol. The human-robot joint alignments and their effects on biomechanical and bio-signal parameters are also studied through musculoskeletal model analysis. The developed predictive model based on the human fatigue and neural network was able to estimate the interaction forces with 80% accuracy. This estimated interaction force in then utilized in the design of an Assist-as-Needed controller.