High-Level Reasoning in Mobile Robotics

 

Deliberative Coordination

This work examines the problem of how a mobile robot may coordinate among multiple, possibly conflicting deliberative processes for reasoning about the physical world. We developed a novel framework by which a mobile robot may learn to coordinate its deliberative reasoning in response to constraints upon processing as well as the performance of each deliberative reasoner. In addition, we demonstrate how the robot may adapt its coordination strategies in response to its experiences using each reasoner.

 

 

 

 

Deliberative Control via Intentional Bias

While at Sony Intelligence Dynamics Laboratory, we developed techniques for incorporating deliberative control within the entertainment robot QRIO. We developed mechanisms for the intentional biasing of activation in lower-level reactive behaviors without subsuming the underlying action selection used to generate natural behavior. Using a structure called the intentional bus, an interface between deliberative and reactive control, we were able to incorporate high-level goals on the system through the modulation of intentional signals sent to the reactive layer.

 

Multi-Robot Systems

Foraging in Heterogeneous Multirobot Teams

In joint work with Tucker Balch, I explored foraging behavior in heterogeneous robot teams. We demonstrated how the team can learn to allocate themselves in an optimal manner given the types of items to be gathered and the relative efficiency of each team member. We then demonstrated that the team allocations learned match those predicted by behavioral ecologists.

 

Mars 20/20

The focus of our research in the DARPA sponsored MARS (Mobile Autonomous Robot Software) 20/20 project lay in the control of heterogeneous teams of unmanned ground and aerial vehicles. In particular, we examined tools, techniques, and behaviors that would enable teams of mobile robots to act and plan in a manner sensitive to the communications constraints present within a particular mission. The focus of my work was on communication recovery, or what the team members should do in the event that communications with each other or a base-station is broken. This work ultimately culminated in a multi-institution joint demonstration at a MOUT (Military Operations in Urban Terrain) site at Fort Benning.

 

 

 

Mission Specification and Task Allocation for Multi-Robot Teams

In joint work with Lockheed-Martin, this project examined techniques for analyzing the usability of multi-robot mission specification software. This line of research culminated in formal usability studies with UAV pilots at the Pax River Naval Air Station. In addition, we examined techniques for integrating case-based mission specification and auction-based task allocation for pre-mission and runtime control of teams of UAVs and UGVs.

 

 

 

 

Biomimetic Formations

As part of the ARO funded HUNT project, I have also worked on developing and testing models of avian leking behavior on multi-robot teams.

 

Intelligent Agents

Model-Based Reflection in Intelligent Software Agents

This project, a collaboration with Ashok Goel and Joshua Jones, has focused on combination of model-based reflection and reinforcement learning within the domain of strategy games. In model-based reflection, an agent contains a model of its own reasoning processes organized via the tasks the agents must accomplish and the knowledge and methods required to accomplish these tasks. Utilizing this self-model, as well as traces of execution, the agent is able to localize failures in its reasoning process and modify its knowledge and reasoning accordingly. We apply this technique to a reinforcement learning problem and show how model-based reflection can be used to locate the portions of the state space over which learning should occur. We performed an experimental investigation of model-based reflection and self-adaptation for an agent performing a specific task (defending a city) in a computer strategy game called FreeCiv.