SUMMARY

As a consequence of the wide adoption of co-bots, the demand for robots that could provide dexterous manipulation in contact-rich and dynamic environments is rising. It could help improve the safety, efficiency, and accuracy of mundane manipulation tasks in application domains such as medical, warehouse, assembly, and biological tissue manipulation. In addition, it could benefit advance other emerging domains such as assistive robotics and rehabilitative robotics. Yet, dexterous contact-rich robotic manipulation presents several challenges, including kinematic dexterity limitations, contact modeling difficulties, adapting to the changing contact parameters, and incorporating contact(e.g., soft material) in trajectory planning. We address these challenges through techniques that span the intersection of kinetic analysis, mechanism design, adaptive force control, and trajectory planning. First, we develop a novel metric to evaluate the dexterity of kinematically constrained (e.g., joint limits, self-collisions) robotic systems. We show its efficacy on 2D redundant chains to generate capability maps, motion generation priors, and motion feasibility inference. Towards developing robots for contact-rich tasks, we show the importance and the relevance of the kinetic (i.e., force) aspect in robotic manipulation. For that, we use a custom-designed force platform to evaluate the kinetics of human trials performed on bi-manual tasks such as suturing and needle insertion. We use it as the basis for incorporating force perception and control in our work. Contact properties vary with the material and surface type. To maintain a stable contact in manipulation tasks while keeping a safe force threshold, an adaptive mechanism is required to probe the changing contact properties and adapt the controller to them. We demonstrate using a custom-designed probe that could be used independently of the robot's configuration to probe the stiffness of the contact in real-time. Then, use the estimated parameters to adapt the admittance force controller to maintain a stable and safe contact throughout the motion. Finally, we demonstrate the importance of using local contact models and the need to incorporate them in trajectory optimization. We show the real-time execution of the framework in Model Predictive Control (MPC) fashion and the capability of the system to undertake bounded motion disturbances.