Bridging the Gap between High-Level Reasoning and Low-Level Control

Abstract

Abstract. We present a formal framework where a nonmonotonic formalism (the action description language C+) is used to provide robots with high-level reasoning, such as planning, in the style of cognitive robotics. In particular, we introduce a novel method that bridges the high-level discrete action planning and the lowlevel continuous behavior by trajectory planning. We show the applicability of this framework on two LEGO MINDSTORMS NXT robots, in an action domain that involves concurrent execution of actions that cannot be serialized.

Combining high-level causal reasoning with low-level geometric reasoning and motion planning for robotic manipulation

“…We describe the preconditions and the effects of the actions as in [3]. For instance, we describe the direct effect of a robot's picking a payload, by the causal laws pick(R) causes holding(R,P) if pickpoint(R)=P.…”

“…Suppose that initially the robots r1 and r2 are at (1,1) and (9,9) respectively; the first payload is located at (3,2) and (8,2); the second payload is located at (8,5) and (3,5); the third payload is located at (9,3) and (9,8). The goal is to move the payloads to the following locations: first payload to (3,2) and (3,7); the second payload to (6,7) and (6,2); the third payload to (9,8) and (9,3). This planning problem can be described in the language of CCALC by means of a "query" as follows:…”

“…With this motivation, [3], [5] introduce an alternative approach to integrating task and motion planning by considering a general interface between them, using "external predicates/functions" whose values for ground instances are computed by an external mechanism, e.g., by a C++ program. The concept of external predicates/functions is not new; they have existed as undocumented features of the planner TLPlan [2] and the Causal Calculator (CCALC) [17].…”

“…The idea in [3], [5] is to use external predicates/functions in the action domain description, for checking the feasibility of a primitive action by a motion planner. [3] applies this approach in the action description language C+ [7] using CCALC, while [5] extends the planning domain description language PDDL [6] to support external predicates/functions (called semantic attachments) and modifies the planner FF [11] accordingly.…”

“…As in [3], [5], we also consider a general interface between high-level causal reasoning and low-level geometric reasoning and motion planning, using external predicates/functions, but in a more flexible way. The first novelty of our approach is the flexible use of external predicates/functions for feasibility checks: Instead of delegating all sorts of feasibility checks to external predicates as in [3], [5], we can decide which sorts of feasibility check should be done by external predicates as part of high-level reasoning, and which sorts of feasibility checks should be done by motion planning later on. For instance, external predicates can check only collisions of robots with each other and with other objects (so they do not check, for instance, collisions of objects with each other), and they can be used in action domain description to specify conditional effects of these robots' actions.…”