Results 1 
4 of
4
Uncertainty and change
 Handbook of Constraint Programming, chapter 21
, 2006
"... Constraint Programming (CP) has proven to be a very successful technique for reasoning about assignment problems, as evidenced by the many applications described elsewhere in this book. Much of its success is due to the simple and elegant underlying formulation: describe the world in terms of decisi ..."
Abstract

Cited by 30 (4 self)
 Add to MetaCart
(Show Context)
Constraint Programming (CP) has proven to be a very successful technique for reasoning about assignment problems, as evidenced by the many applications described elsewhere in this book. Much of its success is due to the simple and elegant underlying formulation: describe the world in terms of decision variables that must be assigned values, place clear and explicit restrictions on the values that may be assigned simultaneously, and then find a set of assignments to all the variables that obeys those restrictions. Thus, CP makes two assumptions about the problems it tackles: 1. There is no uncertainty in the problem definition: each problem has a crisp and complete description. 2. Problems are not dynamic: they do not change between the initial description and the final execution of the solution. Unfortunately, these two assumptions do not hold for many practical and important applications. For example, scheduling production in a factory is, in practice, fundamentally dynamic and uncertain: the full set of jobs to be scheduled is not known in advance, and continues to grow as existing jobs are being completed; machines break down; raw material
An Improved Algorithm for Maintaining Arc Consistency in Dynamic Constraint Satisfaction Problems
 In International Florida AI Research Society Conference
, 2005
"... Real world is dynamic in its nature, so techniques attempting to model the real world should take this dynamicity in consideration. A well known Constraint Satisfaction Problem (CSP) can be extended this way to a so called Dynamic Constraint Satisfaction Problem (DynCSP) that supports adding and rem ..."
Abstract

Cited by 5 (0 self)
 Add to MetaCart
Real world is dynamic in its nature, so techniques attempting to model the real world should take this dynamicity in consideration. A well known Constraint Satisfaction Problem (CSP) can be extended this way to a so called Dynamic Constraint Satisfaction Problem (DynCSP) that supports adding and removing constraints in runtime. As Arc Consistency is one of the major techniques in solving CSPs, its dynamic version is of a particular interest for DynCSPs. This paper presents an improved version of ACDC2 algorithm for maintaining maximal arc consistency after constraint retraction. This improvement leads to runtimes better than the so far fastest dynamic arc consistency algorithm DnAC6 while keeping low memory consumption. Moreover, the proposed algorithm is open in the sense of using either nonoptimal AC3 algorithm keeping a minimal memory consumption or optimal AC3.1 algorithm improving runtime for constraint addition but increasing a memory consumption.
Controlability and Makespan Issues with Robot Action Planning and Execution
, 2006
"... Nowadays, many robotic applications need autonomous decisionmaking capabilities. Among them, some make intensive use of planning. Yet, planning is an activity whose algorithmic complexity is often incompatible with the reactivity ..."
Abstract

Cited by 4 (2 self)
 Add to MetaCart
(Show Context)
Nowadays, many robotic applications need autonomous decisionmaking capabilities. Among them, some make intensive use of planning. Yet, planning is an activity whose algorithmic complexity is often incompatible with the reactivity
Controlability and Makespan Issues with Robot Action Planning and Execution
"... In recent years, there has been an increasing interest in getting planners to “execute ” their plans in the real world. Among them, the temporal “partial order causal link ” planner IxTeT has been endowed with a temporal executive, and has been successfully tested on an ATRV robot. Yet new problems ..."
Abstract
 Add to MetaCart
(Show Context)
In recent years, there has been an increasing interest in getting planners to “execute ” their plans in the real world. Among them, the temporal “partial order causal link ” planner IxTeT has been endowed with a temporal executive, and has been successfully tested on an ATRV robot. Yet new problems arise from the fact that the plan are now being executed. In domains such as robot planning, many tasks have an uncertain duration (a move in a partially unknown environment takes an uncertain amount of time). The current solution uses an interval constraint to represent the possible duration in a Simple Temporal Network. Unfortunately, by doing this, the planner or the executive may reduce the possible durations by propagating other constraints. Hence, this reduction may lead to a plan execution failure. This problem, known as the “controllability” problem, as been addressed by a number of people. For example, the 3DC+ algorithm has been proposed to provide dynamic controllability. Yet to our knowledge, no implementation has been made showing the advantage of such approach. This paper presents the results we got using 3DC+ in IxTeT. Having now a temporal planner which “complies ” with non controllable temporal constraints, we addressed a second problem which arises at execution time. IxTeT is a least commitment planner, and it does not try to minimize the makespan. In fact, the default heuristic is tuned more to spread the actions over the given horizon (to minimize conflict) than to finish the job at the earliest. We implemented a new heuristic which keeps the least commitment strategy yet minimizing the makespan of the plan. This heuristic gives good statistical results on the model we use for our exploration rover type problem. We have tested these two recent modifications on the robot, but also using a very accurate rover simulator. We present real results and simulated one which show the improvement over the previous version/heuristic. We also proposed a solution to a drawback identified during these tests.