No abstract

In this paper, we present a multi-pronged approach to the "Learning from Example" problem. In particular, we present a framework for integrating learning into a standard, hybrid navigation strategy, composed of both plan-based and reactive controllers. Based on the classification of colors and textures as either good or bad, a global map is populated with estimates of preferability in conjunction with the standard obstacle information. Moreover, individual feedback mappings from learned features to learned control actions are introduced as additional behaviors in the behavioral suite. A number of real-world experiments are discussed that illustrate the viability of the proposed method.

No abstract

In this paper, we study hybrid models that not only undergo mode transitions, but also experience changes in dimensions of the state and input spaces. An algorithmic framework for the optimal control of such Multi-Mode, Multi-Dimension (or M 3 D) systems is presented. We moreover derive a detailed M 3 D model for an ice-skater, and demonstrate the use of the developed framework on the ice-skater model.

Multi-modal control is a commonly used design tool for breaking up complex control tasks into sequences of simpler tasks. In this paper, we show that by viewing the control space as a set of such tokenized instructions rather than as real-valued signals, reinforcement learning becomes applicable to continuous-time control systems. In fact, we show how a combination of state-space exploration and multi-modal control converts the original system into a finite state machine, on which Qlearning can be utilized.