Welcome to ActorSim!

Mark “Mak” Roberts mark.roberts.ctr@nrl.navy.mil

Goals unify planning and acting. If a goal is an objective an actor wishes to achieve or maintain, then planning is deliberating on what action(s) best accomplish the objective and acting is deliberating on how to perform each action of a plan. Goal reasoning is deciding what goal(s) to pursue given trade-offs in dynamic, possibly adversarial, environments. Thus, goal reasoning enables more responsive autonomy.

Researchers at the Naval Research Laboratory (Code 5514), led by David Aha and in collaboration with others, have examined numerous goal reasoning topics including early studies in Goal Driven Autonomy, a preliminary formal model, goal formulation, and goal recognition. ActorSim resulted from a need for a general platform for further study. ActorSim is a software platform for studying multiagent goal reasoning in simulated environments. Although goal reasoning has strong ties to the planning, acting, and robotics communities, it is an understudied area of research partly because there exists no publicly available language, definition, and generic implementation. We draw inspiration from the literature in planning, where 15 years of international competitions has blossomed into a research ecosystem of nearly 100 open source planning systems and hundreds of planning benchmarks in a standardized language. Despite this progress, many planning systems focus on a single actor operating within a static environment and static objectives. That is where ActorSim will step in to bridge a needed gap to advance the state of the art.

Broader dissemination of ActorSim will foster deeper study and enriched collaboration between researchers interested in goal reasoning, planning, and acting. ActorSim complements existing open source planning systems with a standardized implementation of goal reasoning so researchers can focus on (1) designing goals and goal transitions for their system (2) linking ActorSim to their particular simulator, and (3) studying goals and behavior in the dynamic environment provided by the simulator. By releasing it as an open source package, we lay a foundation for advanced studies in goal reasoning that include integration with additional simulators and planning systems, formal models, and empirical studies that examine decision making in challenging, dynamic environments.

Component Overview

_images/architecture-summary.png

ActorSim consists of components as shown above.

  • ActorSim Core provides the interfaces and minimal implementations of the platform and contains the essential abstractions that apply across any simulator. This component contains information about Areas, Locations, Actors, Vehicles, Symbols, Maps, Sensors, and configuration details.
  • ActorSim Planning contains the interfaces and minimal implementations for linking to existing open source planning systems. This component unifies Mission Planning, Task Planning, Path Planning, and Motion Planning. The base system includes simple implementations of these planners; we envision linking this component to many open source planning systems.
  • ActorSim Connector links to existing simulators directly or through a network protocol. Currently supported simulators include the MASON simulator of George Mason University and two computer game simulators: StarCraft and Minecraft. We envision links to common robotics simulators (e.g., Gazebo, ROS, OpenAMASE), additional game engines (e.g., Mario Bros., Atari arcade, Angry Birds), and existing competition simulators (e.g., RDDLSim). It may eventually possible to link ActorSim to physical hardware.
  • The Goal Refinement Library provides the abstractions for transitioning goals throughout the system. This library contains the abstractions for goals, goal types, refinements on the goals, goal memory, domain loading, and domain design.
  • ActorSim Coordination houses the interfaces that unify the other components. This component contains abstractions for Tasks, Events, Human interface interaction, Executives, and, event notifications.

Open Source Version

We plan to release the latest ActorSim project as an open source product (most likely through BitBucket) by August 2016 and will link it here. If you are interested in joining the community or the software, please email Mak (mark.roberts.ctr@nrl.navy.mil).

The beta relase of ActorSim (0.6.0) from December 2016 covers the 2015 & 2016 papers for ActorSim as well as some additional features we implemented for early versions of the software. At the moment, it is only available as a 245MB download: ActorSim-0.6.0

To get started:

A web version of the file is available here: Installing the ActorSim Development Environment

Available ActorSim Connectors

The available connectors bundled with the existing source include:

ActorSim Tutorials

A number of ActorSim Tutorials are available for learning ActorSim.

Referencing ActorSim

Roberts, M., Shivashankar, V., Alford, R., Leece, M., Gupta, S., & Aha, D.W. (2016). Goal Reasoning, Planning, and Acting with ActorSim, The Actor Simulator. Proceedings of the Fourth Annual Conference on Advances in Cognitive Systems (to appear). Evanston, IL, USA. PDF

  • An earlier version of this paper appears as: Roberts, M., Alford, R., Shivashankar, V., Leece, M., Gupta, S., & Aha, D.W. (2016). ActorSim: A toolkit for studying Goal Reasoning, Planning, and Acting. In A. Finzi & E. Karpas (Eds.) Planning and Robotics: Papers from the ICAPS Workshop. London, UK. PDF

Indices and tables