Bridging the gap between discrete symbolic planning and optimization-based robot control

Abstract

Symbolic reasoners generate plans which are often not exploiting the robot capabilities and are sensitive to runtime disturbances. This work proposes a scheduler as an interface between a discrete, symbolic plan and a motion control based on constraint optimization. Acting as a local reasoner, the scheduler valuates a set of predicates to decide when an action will be executed. Given a task specification which describes how the action should be realized, the scheduler configures the controller at runtime. A demonstration will be provided considering an “open drawer” scenario

Introducing Geometric Constraint Expressions Into Robot Constrained Motion Specification and Control

“…In our experience, we found out that three priority levels suffice, in most applications (see [13], [23] for use cases). We named these three levels as follows: 1) Safety constraints: constraints that are necessary for the robot platform or critical surroundings integrity, (e.g.…”

“…This allows us to decouple the language-dependent task grounding from the formal description, not only to guarantee portability and legibility, but also to exploit existing tools and model validators. The validation of the model semantic is the first step to allow automatic reasoning systems that generate symbolic plans, [11], [12], or symbolic task scheduling, [13]. Further numerical post-check operations have been implemented in the host language of the controller.…”

“…Monitors can be used to verify pre-, per-as well as post-conditions, e.g. [13]. A monitor can either observe: 1) a controllable variable (for example a variable that is used as constraint), or 2) a measured quantity that is influenced by the robot actions, but cannot be directly controlled.…”

“…Having a well defined end-of-task monitor is instrumental to define the post-condition(s) of the task to be used in plan reasoning and scheduling algorithm (see [13] and references). …”

“…The benefit of this process is manifold: by defining constraints with a limited number of formally described elements, it becomes possible to design reasoning algorithms to sort out the best plan for a given action (e.g. [11], [13]) and check (semantically) whether a task is ill-or welldefined, without prior knowledge of the solver specifics, and/or with a lot less involvement of human developers. This progress helps in the "ease of use" of task specification, but also in "verification & validation" of robotic applications.…”