Human Interaction and Robotics Group (HIRO) project repository maintained and developed by Ava Abderezaei, Caleb Kumar and Noah Liska.
Planning trajectories under complex dynamic constraints is computationally expensive and time-consuming. Hence, employing these dynamic constraints in motion planning, renders practical real-time motion planning unachievable. In this project, we propose a method to generate trajectories for systems with complex dynamics without explicitly using the dynamics. We specifically focus on the problem of robot-assisted feeding using a spoon. To that end, we generate spill-free motion primitives and then use these primitive motions to generate a spill-free trajectory. Additionally, by implementing our proposed method on top of an optimization solver, it can work in real-time. This is specifically important in assisted-feeding settings, as the robot should be able to react to environmental changes in real-time. To that end, we formalize a safe zone for the robot to enter when the user is distracted during the feeding process. We evaluate our approach on a panda robotic arm and achieve a spill-free trajectory over different foods including sugar, peas, and yogurt.
