Résumé
The surge in e-commerce has outpaced the capacity of human labor, which has long been the backbone of logistics. This pressing issue necessitates the integration of robots into repetitive and labor-intensive tasks, marking a significant shift in the logistics landscape.Humans employ diverse parcel-handling techniques, ranging from pre-grabbing maneuvers to placing and throwing. In contrast, robots have been constrained in their ability to pick and place. This thesis introduces an innovative approach to address this disparity by iterating impact-aware control with model-based task definitions.The first part of this thesis details a novel strategy for planning robotic motions tailored to impact-intensive tasks such as tossing, grabbing, and boxing. While the current state of robotics technology features high-performance hardware, robots tend to decelerate to zero velocity when approaching contact (e.g., for picking up items).Our research focuses on enabling motion generation to embrace contact at non-zero velocity configurations and facilitate non-stop trajectory cycles. This is achieved by enhancing a sampling-based planner that integrates kinodynamic constraints and utilizes kinematic redundancy to define optimized paths. We also explore high-dynamic logistic tasks —like tossing items to different targets with or without controlled impact velocity, dual-arm grabbing, and dynamic box tilting—applying the improved planner to these scenarios. This approach has demonstrated efficiency in both motion generation and the robot's control ability to handle impacts and perform complex maneuvers effectively.Following this, the thesis presents a model-based method for determining the optimal sequence of pokes to achieve desired planar motions using a robotic manipulator. Nonprehensible manipulations, where grasping is not involved, have proven valuable in real-world applications. One area of interest is manipulating objects on a planar environment. However, this task could be expedited by introducing impacts into the process. A new mixed-integer nonlinear problem is formulated to tackle this challenge, integrating a carefully designed planar friction model that accounts for both sliding and spinning motions. The optimized poking actions are then used to compute the robot's end-effector velocities and directional effective inertia. Afterward, an impact-aware jerk-based planner and quadratic programming (QP) are employed to plan and track the desired impact trajectories.Lastly, the thesis explores trajectory optimization taking into account the modeling of suction cup deformation in the different phases of manipulation (including grabbing, holding, and tossing). This is motivated mainly by observations during preliminary experiments and results.A trajectory optimization is formulated to handle the discussed suction cup deformable model. This investigation improves our understanding of how suction cups deform under high acceleration motions, ultimately enabling efficient planning for tossing tasks.In conclusion, this thesis aims to improve robots' capabilities to handle impacts through impact-aware planning techniques and detailed task modeling, ultimately enhancing their operational efficiency in logistics applications. This work contributes to the broader goal of developing more adaptable and resilient robotic systems for the industrial sector.