James P. Gunderson

An assistive robotic agent for pedestrian mobility

The goal of this project is to develop a pedestrian mobility aid for the elderly. In order for this type of assistive technology to be useful and accepted by its intended user community, it must enhance the abilities of users, not replace them. This leads to an agent architecture in which the agent must operate without hindering the users ability to take direct action when they choose. In other words, the agent cannot simply be a proxy for the users actions. The agent must select its own goals based on observations of its users actions. This is crucial not only because users may have diminished capacity to explain their actions to an agent, but because the ability of the agent to correctly interpret the users goals is tied to its ability to act while still allowing the user to “feel in control”. We present a mobility aid, i. e. a wheeled walker, which varies its goals and level of activity based on an estimation of its users intentions. The assistive agent often takes no action, allowing the user to be fully in control. When the ease or safety of the users travel is threatened, the agent attempts to influence the users motion based on its belief in the users goal. By varying the degree of autonomy, the walker can adjust to the user as their abilities change from day to day, or hour to hour. This prevents the walker from trying to do too much, allowing the user to feel as if they are in control and not being lead.

Adaptive goal prioritization by agents in dynamic environments

We present a model for an agent architecture that supports adaptive goal prioritization. Goal prioritization is the ability of the agent to adjust the relative priorities of a goal set in response to changes in a dynamic environment. This architecture allows an agent to act pro-actively to achieve goals, and to dynamically determine which goals should be pursued at any time. We discuss a dynamic domain in which the architecture will be situated, and outline the techniques needed to support goal re-prioritization.

The effects of uncertainty on plan success in a simulated maintenance robot domain

Autonomous robots must be designed to function in real-world domains. These domains are characterized by uncertainty, yet little work has been done to quantify the effects of uncertainty on goal satisfaction. Our research on simulated repair robots indicates that different types of uncertainty have fundamentally different relationships to plan success. This paper investigates the impacts of sensor inaccuracies, exogenous events and failures in both motion and gripper operations. Extensive simulations provide the basis for mathematical models of the relationship between these types of uncertainties and plan success. In addition, the impact of current heuristics such as ‘retry on failure’, and ‘tell me three times’ is investigated. Concluding that even low levels of uncertainty can significantly impact goal satisfaction in a stimulated repair robot domain and that repeated attempts provide significant improvement in plan success rates in the face of uncertainty.

A linear time transform for probability aware planning

Presents a transform that enables traditional shortest-feasible-plan planners to reason about uncertain operators and produce plans which have higher probabilities of success. This transform converts a probability-aware domain description into a STRIPS-style description, where the probability of success is expressed by the plan length. Using this transformed description, a plan can be generated by a traditional planner. The transform is shown to be at worst linear in the size of the input, and allows the planning system to trade-off accuracy against runtime as an anytime computation.