In the context of movement assistance, where physical contact between robotic systems and the human body is expected to increase, the long-term physical health of patients must be investigated. This is precisely one of the commercial arguments promoting assistive devices, which are claimed to limit the impact of traumatic tasks on health or to bridge the gap between disabled and able-bodied patients. Paradoxically, there is little to no consideration of the human body biomechanics in the control of these devices. The reasons for such a shortcoming are twofold: i) the complexity and lack of fidelity of neuromusculoskeletal (MSK) models and ii) the slowness of algorithms relying on these models which disqualify them for real-time applications, such as typically needed for controlling assistive devices. B-IRD stands at a crossroad between robotics and biomechanics to develop faithful and personalized MSK models and fast algorithms that will enable to take into account the complex dynamics of the human body in the control of assistive devices. The ambition of the first axis is to develop biomechanical estimation methods, fast enough to be used in feedback for the control of assistive devices, without sacrificing their accuracy. The objective of the second axis is to develop MSK personalization methods which exploit the high kinematic accuracy of robots. The last axis presents the integration of these algorithms and models in the control of assistive devices to generate a customized motion assistance that truly respects the biomechanical limits of the human body while exploiting its assets. To improve the benefits of assistive technologies, B-IRD will enable to objectively quantify the quality of their assistance and to change their control paradigm. The foreseen approaches are based on robotics, automatic control and machine learning tools, not commonly used in biomechanics.