In contrast to humans who use many types of manipulation to accomplish daily tasks and can easily sequence and execute pertinent actions, robots are confined to simple tasks that are often painstakingly broken down by the humans who operate those robots. Towards the goal of increasing robot autonomy, this project will extend the capabilities of manipulation planners. For the purposes of this research, manipulation planning is the domain in-between classical motion planning and what is often called task and motion planning, which includes temporal reasoning and high-order logics. This research adopts a constraint-centric view and defines a set of low-dimensional subspaces, or modes, amongst which the system must transition. The definition of transitions is also constraint-centric and is only possible because of the unified approach used when considering modes. The work depends on constructs from differential geometry and the use of powerful motion planners. It adopts a synergistic layered scheme where a discrete planner decides the sequence of modes while being constantly informed by a continuous planner that attempts the transitions between modes. The work will start with a specific but general type of constraints, manifold constraints, and later expand to other types. The proposed research will identify the limits of using constraints as a unifying construct in manipulation planning and in doing so, it will also allow for the incorporation of manipulation-specific primitives that can extend the framework.
This work has been supported by grant NSF RI 2008720.