Elly Winner

ICML Conference 2003 Conference Paper

DISTILL: Learning Domain-Specific Planners by Example

• Elly Winner
• Manuela Veloso

ICAPS Conference 2002 Conference Paper

Analyzing Plans with Conditional Effects

• Elly Winner
• Manuela Veloso

Several tasks, such as plan reuse and agent modeling, rely on interpreting a given or observed plan to generate the underlying plan rationale. Although there are several previous methods that successfully extract plan rationales, they do not apply to complex plans, in particular to plans with actions that have conditional effects. In this paper, we introduce SPRAWL, an algorithm to find a minimal annotated partially ordered structure that maximizes a given evaluation function for an observed totally ordered plan with conditional effects. The algorithm proceeds in a two-phased approach, first preprocessing the given plan using a novel needs analysis technique that builds a needs tree to identify the dependencies between operators in the totally ordered plan. The needs tree is then processed to construct a partial ordering that captures the complete rationale of the given plan. We also provide a polynomial-time algorithm to find non-optimal minimal annotated partial orderings of observed totally ordered plans with conditional effects. We provide illustrative examples and discuss the challenges we faced.

ICAPS Conference 2000 Conference Paper

Vision-Servoed Localization and Behavior-Based Planning for an Autonomous Quadruped Legged Robot

• Manuela Veloso
• Elly Winner
• Scott Lenser
• James Bruce
• Tucker R. Balch

Planning actions for real robots in dynamic, and mlcertain environments is a challenging problem. It is not viable to use a complete model of the world: it is most appropriate to achieve goals mid handle uncertainty by integrating deliberation and behavior-based reactive planning. Wesuccessfully developed a system integrating perception and action for the RoboCup99 Sony legged robot league. The quadruped legged robots are fully autonomous and thus must. have onboard vision, localization a~td action selection. We briefly present our perception algorithm that automatically classifies arid tracks colored blobs in real time. We then briefly introduce our Sensor Resetting I, ocalization (SRL) algorithm which is an extension of Monte Carlo Localization. Vision and localization provide the state input for action selection. Our robust and sensible behavior scheme handles dynamic changes in information accuracy. We developed a utility-based system for using mid acquiring location information. Finally, we have devised several special built-in plans to deal with times when urgent action is needed and the robot cannot afford to colh: ct accurate location information. Wepresent results using the real robots, which demonst. rate the success of our approach. Our team of Sony quadruped legged robots, CMTrio-99, won all but one of its games in RoboCup99, and was awarded third place in the competition.