Bayesian Multi-Task Learning MPC for Robotic Mobile Manipulation

Mobile manipulation in robotics is challenging due to the need solve many diverse tasks, such as opening a door or picking-and-placing an object. Typically, basic first-principles system description of robot available, thus motivating use model-based controllers. However, dynamics and its interaction with object are affected by uncertainty, limiting controller's performance. To tackle this problem, we propose Bayesian multi-task learning model that uses trigonometric basis functions identify error dynamics. In way, data from different but related tasks can be leveraged provide descriptive efficiently updated online for new, unseen tasks. We combine scheme predictive controller, extensively test effectiveness proposed approach, including comparisons available baseline present simulation tests ball-balancing robot, hardware experiments quadrupedal manipulator.