Toward Obstacle Avoidance in Intermittent Dynamical Environments

In this paper we discuss a robotic task requiring dynamical safety in the face of an intermittent environment. We define and offer examples of this notion. We then construct a dynamically safe composite controller from dynamically safe constituents, and present empirical evidence of its effectiveness. 1. I n t r o d u c t i o n This paper and a companion paper [1] develop an approach to building up complexes of controllers from simpler constituents. These ideas, first set out in [2], arise from our previous experience with dynamically dexterous robot tasks. The composite controllers we develop result in more varied and capable closed loop robot behavior than can be accomplished by any one of the constituent controllers acting in isolation. The companion paper [1] presents a method for extending the domain of attraction of a goal set via an appropriate composition procedure. This paper introduces the notion of dynamical safety for systems with only intermittent contact between robot and environment. In other words, we seek to add obstacle avoidance capability to robot controllers that can already successfully manipulate objects that must be periodically released into a dynamical environment along the way toward their goal. Although all the discussion in the companion paper is limited to the dynamical setting, we believe that our framework for controller composition is useful over a much wider class of robot task domains. In contrast, the novelty of the obstacle avoidance problem considered here is peculiar to the intermittent setting. To introduce the new notion of a safe control strategy we find it useful to add to the formal definition (i) a specific extended example along with (ii) some empirical results reflecting our efforts to implement the example on our three degree of freedom B/ihgler robot [3].