Search

Space exploration applications can benefit greatly from autonomous systems. Great distances, limited communications and high costs make direct operations impossible while mandating operations reliability and efficiency beyond what traditional commanding can provide. Autonomous systems can improve reliability and enhance spacecraft capability significantly. However, there is reluctance to utilizing autonomous systems. In part this is due to general hesitation about new technologies, but a more tangible concern is that of reliability of predictability of autonomous software. In this paper, we describe ongoing work aimed at increasing robustness and predictability of autonomous software, with the ultimate goal of building trust in such systems. The work combines state-of-the-art technologies and capabilities in autonomous systems with advanced validation and synthesis techniques. The focus of this paper is on the autonomous system architecture that has been defined, and on how it enables the application of validation techniques for resulting autonomous systems.

LORAX is a robotic astrobiological study of the ice field surrounding the Carapace Nunatak near the Allan Hills in Antarctica. The study culminates in a l00km traverse, sampling the ice at various depths (from surface to 10cm) at over 100 sites to survey microbial ecology and to record environmental parameters. The autonomy requirements from LORAX are shared by many robotic exploration tasks. Consequently, the LORAX autonomy architecture is a general architecture for on-board planning and execution in environments where science return is to be maximized against resource limitations and other constraints.

NASA spacecraft and robots operate at long distances from Earth Command sequences generated manually, or by automated planners on Earth, must eventually be executed autonomously onboard the spacecraft or robot. Software systems that execute commands onboard are known variously as execution systems, virtual machines, or sequence engines. Every robotic system requires some sort of execution system, but the level of autonomy and type of control they are designed for varies greatly. This paper presents a survey of execution systems with a focus on systems relevant to NASA missions.

Operating the Mars Exploration Rovers is a challenging, time-pressured task. Each day, the operations team must generate a new plan describing the rover activities for the next day. These plans must abide by resource limitations, safety rules, and temporal constraints. The objective is to achieve as much science as possible, choosing from a set of observation requests that oversubscribe rover resources. In order to accomplish this objective, given the short amount of planning time available, the MAPGEN (Mixed-initiative Activity Plan GENerator) system was made a mission-critical part of the ground operations system. MAPGEN is a mixed-initiative system that employs automated constraint-based planning, scheduling, and temporal reasoning to assist operations staff in generating the daily activity plans. This paper describes the adaptation of constraint-based planning and temporal reasoning to a mixed-initiative setting and the key technical solutions developed for the mission deployment of MAPGEN.

In this paper we provide a formal framework to define the scope of planning agents based on a single declarative model. Having multiple agents sharing a single model provides numerous advantages that lead to reduced development costs and increase reliability of the system. We formally define planning in terms of extensions of an initial partial plan, and a set of flaws that make the plan unacceptable. A Flaw Filter (FF) allows us to identify those flaws relevant to an agent. Flaw filters motivate the Plan Identification Function (PIF), which specifies when an agent is is ready hand control to another agent for further work. PIFs define a set of plan extensions that can be generated from a model and a plan request. FFs and PIFs can be used to define the scope of agents without changing the model. We describe an implementation of PIFsand FFswithin the context of EUROPA, a constraint-based planning architecture, and show how it can be used to easily design many different agents.

Planning and scheduling in the face of uncertainty and change pushes the capabilities of both planning and dialogue technologies by requiring complex negotiation to arrive at a workable plan. Planning for use of semi-autonomous robots involves negotiation among multiple participants with competing scientific and engineering goals to co-construct a complex plan. In NASA applications this plan construction is done under severe time pressure so having a dialogue interface to the plan construction tools can aid rapid completion of the process. But, this will put significant demands on spoken dialogue technology, particularly in the areas of dialogue management and generation. The dialogue interface will need to be able to handle the complex dialogue strategies that occur in negotiation dialogues, including hypotheticals and revisions, and the generation component will require an ability to summarize complex plans. This demonstration will describe a work in progress towards building a spoken dialogue interface to the EUROPA planner for the purposes of planning and scheduling the activities of a semi-autonomous robot. A prototype interface has been built for planning the schedule of the Personal Satellite Assistant (PSA), a mobile robot designed for micro-gravity environments that is intended for use on the Space Shuttle and International Space Station. The spoken dialogue interface gives the user the capability to ask for a description of the plan, ask specific questions about the plan, and update or modify the plan. We anticipate that a spoken dialogue interface to the planner will provide a natural augmentation or alternative to the visualization interface, in situations in which the user needs very targeted information about the plan, in situations where natural language can express complex ideas more concisely than GUI actions, or in situations in which a graphical user interface is not appropriate.

Planning research in Artificial Intelligence (AI) has often focused on problems where there are cascading levels of action choice and complex interactions between actions. In contrast. Scheduling research has focused on much larger problems where there is little action choice, but the resulting ordering problem is hard. In this paper, we give an overview of M planning and scheduling techniques, focusing on their similarities, differences, and limitations. We also argue that many difficult practical problems lie somewhere between planning and scheduling, and that neither area has the right set of tools for solving these vexing problems.

Tractable classes of constraint satisfaction problems are of great importance in artificial intelligence. Identifying and taking advantage of such classes can significantly speed up constraint problem solving. In addition, tractable classes are utilized in applications where strict worst-case performance guarantees are required, such as constraint-based plan execution. In this work, we present a formal framework for search-free (backtrack-free) constraint satisfaction. The framework is based on general procedures, rather than specific propagation techniques, and thus generalizes existing techniques in this area. We also relate search-free problem solving to the notion of decision sets and use the result to provide a constructive criterion that is sufficient to guarantee search-free problem solving.

Many complex real-world decision and control problems contain an underlying constraint reasoning problem. This is particularly evident in a recently developed approach to planning, where almost all planning decisions are represented by constrained variables. This translates a significant part of the planning problem into a constraint network whose consistency determines the validity of the plan candidate. Since higher-level choices about control actions can add or remove variables and constraints, the underlying constraint network is invariably highly dynamic. Arbitrary domain-dependent constraints may be added to the constraint network and the constraint reasoning mechanism must be able to handle such constraints effectively. Additionally, real problems often require handling constraints over continuous variables. These requirements present a number of significant challenges for a constraint reasoning mechanism. In this paper, we introduce a general framework for handling dynamic constraint networks with real-valued variables, by using procedures to represent and effectively reason about general constraints. The framework is based on a sound theoretical foundation, and can be proven to be sound and complete under well-defined conditions. Furthermore, the framework provides hybrid reasoning capabilities, as alternative solution methods like mathematical programming can be incorporated into the framework, in the form of procedures.