Randomization in Robot Tasks

Abstract

This paper explores the role of randomization in the solution of robot manipulation tasks. Randomization refers to the random selection and execution of an action from a collection of possible actions. The intention is that this collection contains some actions that are useful for making progress towards accomplishing a task, but that the precise identity of these useful actions is unknown. Randomization offers one approach for ensuring progress in a probabilistic sense. An example of randomization is given by the strategy of shaking a bin containing a part in order to orient the part in a desired stable state with some high probability. Another example consists of using reliable sensory information to bring two parts close together, then relying on short random motions to actually mate the two parts, once the part motions lie below the available sensing resolution. Further examples include tapping parts that are tightly wedged, twirling gears before trying to mesh them, and vibrating parts to facilitate a mating operation. Randomization is also useful for mobile robot navigation and as a means of guiding the design process. Randomization is useful in three basic ways. First, randomization can increase the class of solvable tasks. This is because a randomized strategy need not guarantee task success with certainty in a specific number of steps, relying instead on repeated execution of the randomizing actions to accomplish the task in an expected sense. Second, randomization can reduce a strategy's knowledge requirements. This is because randomization can tolerate and circumvent local failures, thereby making a strategy less sensitive to task details. Third, randomization can simplify the planning and execution process. This is because a randomized solution may be able to ignore precise prediction of special case scenarios, instead simply ensuring eventual accomplishment of the task independent of the actual scenarios encountered.