We introduce a generative system for what we call deflection drawings, drawings made by virtual drawing robots, or drawbots, executing a random walk based on a curvature model of Chappell. Deflection drawings are characterized by the property that drawbots simultaneously avoid the paths of other robots as well as a small number of symmetrically placed obstacles. We formulate a fitness function for aesthetically evaluating such drawings and an evolutionary framework for evolving them. We provide examples of evolved drawings and offer an in-depth analysis of one particular drawing. Finally, we consider techniques for validating our evolutionary framework by considering the influence of previously formulated principles that apply to generative systems that are highly stochastic in nature.