William T. B. Uther

Vision, Strategy, and Localization Using the Sony Robots at RoboCup-98

Sony has provided a robot platform for research and development in physical agents, namely, fully autonomous legged robots. In this article, we describe our work using Sonys legged robots to participate at the RoboCup-98 legged robot demonstration and competition. Robotic soccer represents a challenging environment for research in systems with multiple robots that need to achieve concrete objectives, particularly in the presence of an adversary. Furthermore, RoboCup offers an excellent opportunity for robot entertainment. We introduce the RoboCup context and briefly present Sonys legged robot. We developed a vision-based navigation and a Bayesian localization algorithm. Team strategy is achieved through predefined behaviors and learning by instruction.

TTree: tree-based state generalization with temporally abstract actions

In this chapter we describe the Trajectory Tree, or TTree, algorithm. TTree uses a small set of supplied policies to help solve a Semi-Markov Decision Problem (SMDP). The algorithm uses a learned tree based discretization of the state space as an abstract state description and both user supplied and auto-generated policies as temporally abstract actions. It uses a generative model of the world to sample the transition function for the abstract SMDP defined by those state and temporal abstractions, and then finds a policy for that abstract SMDP. This policy for the abstract SMDP can then be mapped back to a policy for the base SMDP, solving the supplied problem. In this chapter we present the TTree algorithm and give empirical comparisons to other SMDP algorithms showing its effectiveness.

The Lumberjack Algorithm for Learning Linked Decision Forests

While the decision tree is an effective representation that has been used in many domains, a tree can often encode a concept inefficiently. This happens when the tree has to represent a subconcept multiple times in different parts of the tree. In this paper we introduce a new representation based on trees, the linked decision forest, that does not need to repeat internal structure. We also introduce the Lumberjack algorithm for growing these forests in a supervised learning setting. Lumberjack induces new subconcepts from repeated internal structure. This allows Lumberjack to represent many concepts more efficiently than a normal tree structure. We then show empirically that Lumberjack improves generalization accuracy on these hierarchically decomposable concepts.

The lumberjack algorithm for learning linked decision forests

While the decision tree is an effective representation that has been used in many domains, a tree can often encode a concept inefficiently. This happens when the tree has to represent a subconcept multiple times in different parts of the tree. In this paper we introduce a new representation based on trees, the linked decision forest, that does not need to repeat internal structure. We also introduce a supervised learning algorithm. Lumberjack, that uses the new representation. We then show empirically that Lumberjack improves generalization accuracy on hierarchically decomposable concepts.