Scope of the Workshop
Multi-objective motion generation with highly redundant robots have been studied extensively over the last decades. Multiple torque- or velocity-control modules can be synthesized to achieve a range of control objectives simultaneously. Prioritization schemes are one of the preferred means to deal with potentially conflicting control objectives, that represent tasks and constraints.
The aspect of priority management and scheduling is important for future robotic applications in dynamic, unstructured and unpredictable or uncertain environments. However, the choice of priorities and continuous priority rearrangement is non-trivial, especially when requiring strict decoupling of objectives through nullspace projections.
Related works that extend the classical stack-of-tasks formulation often restrict priority swapping operations to consecutive levels in the stack and/or allow task-insertion and -removal only at the lowest priority level. Many of these approaches are computationally inefficient, especially for multiple simultaneous tasks in transition. In spite of significant advances in recent years, current systems are generally far from adequate for actual field deployments. Therefore, the questions we want to address:
• How to remove/add tasks or change the priorities of tasks in any prioritization scheme (soft / strict / hybrid) to cope with dynamically changing contexts or modified high-level goals?
• How can robot priorities be learned based upon human-demonstrations or cost-functions?
• What are appropriate planning algorithms that include priority management and scheduling?
• What are the advantages and disadvantages of initiating priority transitions at high-level planning stage compared to low-level controller decisions?
• What are the challenges in achieving the proposed goal in different applications (e.g. industrial human-robot collaboration, household service robots, teleoperated semi-autonomous robots, etc.)?