Real-Time Simulation of Robotic Hand Prosthesis and Validation of Object-Interaction Dynamics Modeling

To provide adequate functionality, an upper limb prosthesis must be sufficiently dexterous and capable of fast, precise responses to the user’s inputs, while still generating enough torque to securely grasp objects. These design objectives are often conflicting, requiring careful tuning of both the mechanical properties of the device and its control strategies. Simulating components in isolation frequently produces inconsistencies once they are integrated, highlighting the importance of a comprehensive system-level simulation. The static, kinematic, and dynamic behaviors of the prosthesis must therefore be modeled to establish a reliable foundation for improving design and control without the need to construct physical prototypes. This digital twin approach, already widely adopted across industrial sectors, presents particular challenges in the context of upper limb prostheses. The need for rapid responses, the central role of object contact, and the lightweight nature of the structures make dynamic simulations computationally demanding and prone to instability. These factors limit the effectiveness of standard physics engines for accurate reproduction of system behavior and hinder their use as dependable design tools in this field. To mitigate these difficulties, most existing simulation models of similar devices adopt simplifications that reduce computational cost at the expense of fidelity to the real system. This thesis focuses on the development of a digital twin of the robotic prosthetic platform DexterHand, a project carried out within the Rehab Technologies Lab at IIT (Italian Institute of Technology) and funded by INAIL (Italian National Institute for Insurance against Accidents at Work). The model was designed with a strong focus on accuracy, avoiding the introduction of fictitious physical elements that are often used to reduce the computational cost of the simulation (e.g. false moments of inertia to stabilize the joints), but do not find correspondence in the real environment. Therefore, avoiding the implementation of such measures ensures the reliability of the simulation. The result is a simulation interface that reproduces the operation of the physical prototype and makes it possible to test control algorithms (a defining feature of DexterHand, and thus of the model, being the independent control of individual fingers) as well as to encode feedback to be delivered to the patient according to the dynamics of contact with diverse objects. The final model, which closely replicates the physical behavior of the prototype, enables the rapid development and assessment of hardware and software solutions without their immediate implementation on the real device. It also allows their impact on the overall system to be studied without relying on isolated component simulations. The use of MuJoCo, a state-of-the-art physics engine, combined with Unity, a versatile development environment well-suited for integration with multiple external devices, enables the creation of a model that not only accurately reflects the dynamics of a physical prototype but also supports a fast setup of experimental tests using a wide range of peripherals.