Doctoral Research Seminar on "Task and motion planning using object-centered predicates and action contexts" by Dr. Alejandro Agostini, HCR/TUM.
Task and motion planning (TAMP) frameworks use an abstract representation of physical changes with actions, encoded into planning operators, to quickly define sequences of (symbolic) actions for the robot to achieve a desired configuration of objects (or goal) from variable initial situations. In this manner, well-studied robotic actions such as pushing, picking, or placing can be concatenated for the execution of multi-step tasks. One of the main challenges of TAMP is that domain- and task-specific physical constraints that are important for defining realizable plans are difficult to assess with the abstract representation of task planners. This assessment is normally carried out outside task planning using expensive geometric reasoning and multiple calls on unfeasible trajectories.
We propose an abstract domain representation that consistently encodes relevant geometric constraints to favour the generation of physically feasible plans. These constraints are described using an object-centered perspective that can be automatically linked to robot sensing and acting parameters (e.g. object poses and DMPs) without handcrafting signal-symbol relations. For action execution, the approach evaluates the intention of an action in the context of the task: pick a bottle for pouring afterwards, to generate trajectories considering motion dependencies between consecutive actions.
June 7th, 11h online via zoom. On how to participate, please look on the teaching website: https://www.ei.tum.de/ics/teaching/sose2021/research-seminar-for-doctoral-candidates/