We seek a postdoctoral researcher (USA citizenship required) to collaborate on the topic of goal reasoning, which concerns the capability of an agent to (within constraints) self-govern its behavior (e.g., the monitoring and self-selection of goals). Our new project focuses on developing a Tactical Battle Manager (TBM) to enable teaming of simulated manned and unmanned aircraft. The TBM must plan and coordinate actions of multiple aircraft in a highly dynamic, uncertain, and adversarial simulation environment. NRL's focus will be on the use of goal reasoning in a multi-layer hierarchical decision architecture in which higher levels, in accordance with human direction, determine goals for the autonomous systems (while lower levels coordinate and optimize vehicle actions automatically). The TBM will be integrated into high-fidelity mission simulators that are being used for air combat analysis, training, and technology assessment.
Requirements include research experience on topics related to artificial intelligence (e.g., multiagent systems, planning, machine learning), software development experience, and (preferably) the use of simulations in research studies. The individual(s) should be able to design, develop, integrate, and evaluate goal reasoning models/algorithms in a decision making system (i.e., TBM), and report on the analysis of associated research hypotheses. While beneficial, no prior expertise on the application domain is required. NRL's post-doctoral research programs are described below.
Keywords: Decision making, goal reasoning, multiagent systems, modeling, planning, machine learning, simulation, autonomous air vehicles
We seek a post-doctoral researcher to join us on a project (starting Fall 2013) on designing/developing/testing a situated decision-making process in which agents can autonomously learn and pursue their objectives (i.e., to respond competently to unanticipated states without requiring exhaustive pre-programming), and its application to controlling distributed mobile sensors in a variety of ISR mission scenarios. The ideal candidate (must be a US citizen or permanent resident) would have some familiarity with a specification language (e.g., using Linear Temporal Logic (LTL)) for distributed robotic control and a suitable robotics architecture (e.g., ROS, MOOS). The following paragraphs provide more detail.
Context awareness is a key factor in the success of autonomous systems, as any planning or control algorithm makes a number of assumptions about the autonomous agent and its environment. Planners encode these assumptions as predicates before executing a task, and lower-level control schema are generally designed to use local sensing and decision making to enhance robustness. When the predicates are violated or surprises emerge, replanning algorithms such as Goal Driven Autonomy (GDA) enable the agent to replan based on its current context. Recent work in both biologically inspired swarming and potential field based team control has focused on reactive finite state autonoma (RFSA), in which agents interact with their environments using different rules depending on their sensed context. However, the definition of "context" and "sensing" have generally been ad hoc and platform- or environment-specific.
Linear temporal logic (LTL) provides a provably correct means of synthesizing RFSA's based on specifications generated by a person or agent. It provides a precise definition of both context and sensing for mode switching within an RFSA and a robust means of reporting surprises (e.g., to a GDA agent). Our project consists of designing an implementation of LTL that can take a plan generated by a high level planner (e.g., SHOP2) and express it, to use Kress-Gazit et al.'s terminology, in terms of objectives, actions and sensors for a team of small mobile robots. We want to enable fully autonomous team operations in permissive, contested, and austere environments.
Our facilities include 10 Ascending Technologies Pelican quadrotors, several ARDrone2 quadrotors, 3 PackBots, and 8 Pioneer3AT autonomous vehicles to use for experiments as well as access to the DoD's High Performance Computing Modernization Program (HPCMP) for computationally intense tasks such as plan optimization and simulations. We expect to mount a variety of novel radio frequency (RF), chemical, radiological and visual sensors on our platforms and use them to find and map plumes of hazardous materials, sources of interesting signals and other objects.
If these opportunities interest you and you qualify, please contact me to discuss them.
Last updated: 10 September 2013