Learning to Repair Plans and Schedules Using a Relational (deictic) Representation
نویسندگان
چکیده
Unplanned and abnormal events may have a significant impact on the feasibility of plans and schedules which requires to repair them ‘on-the-fly’ to guarantee due date compliance of orders-in-progress and negotiating delivery conditions for new orders. In this work, a repair-based rescheduling approach based on the integration of intensive simulations with logical and relational reinforcement learning is proposed. Based on a relational (deictic) representation of schedule states, a number of repair operators have been designed to guide the search towards a goal state. The knowledge generated via simulation is encoded in a relational regression tree for the Q-value function defining the utility of applying a given repair operator at a given schedule state. A prototype implementation in Prolog language is discussed using a representative example of three batch extruders processing orders for four different products. The learning curve for the problem of inserting a new order vividly illustrates the advantages of logical and relational learning in rescheduling.
منابع مشابه
SmartGantt - An intelligent system for real time rescheduling based on relational reinforcement learning
With the current trend towards cognitive manufacturing systems to deal with unforeseen events and disturbances that constantly demand real-time repair decisions, learning/reasoning skills and interactive capabilities are important functionalities for rescheduling a shop-floor on the fly taking into account several objectives and goal states. In this work, the automatic generation and update thr...
متن کاملTask Rescheduling using Relational Reinforcement Learning
Generating and representing knowledge about heuristics for repair-based scheduling is a key issue in any rescheduling strategy to deal with unforeseen events and disturbances. Resorting to a feature-based propositional representation of schedule states is very inefficient and generalization to unseen states is highly unreliable whereas knowledge transfer to similar scheduling domains is difficu...
متن کاملLearning knowledge-level domain dynamics
The ability to learn relational action models from noisy, incomplete observations is essential to support planning and decision-making in real-world environments. While some methods exist to learn models of STRIPS domains in this setting, these approaches do not support learning of actions at the knowledge level. In contrast, planning at the knowledge level has been explored and in some domains...
متن کاملThe Thing that we Tried Didn't Work very Well: Deictic Representation in Reinforcement Learning
Most reinforcement learning methods oper ate on propositional representations of the world state. Such representations are of ten intractably large and generalize poorly. Using a deictic representation is believed to be a viable alternative: they promise gener alization while allowing the use of existing reinforcement-learning methods. Yet, there are few experiments on learning with deic ti...
متن کاملDeictic Option Schemas
Deictic representation is a representational paradigm, based on selective attention and pointers, that allows an agent to learn and reason about rich complex environments. In this article we present a hierarchical reinforcement learning framework that employs aspects of deictic representation. We also present a Bayesian algorithm for learning the correct representation for a given sub-problem a...
متن کامل