Recognizing objects from visual percepts is a challenging task for mobile robots. Conventional approaches for object recognition are tour de force, relying almost exclusively on complex statistical models for classification and heavy engineering of computationally-intensive features [4, 14]. The processing capability of the recognition system, however, is seldom questioned and assumed to be unconstrained. In stark contrast, mobile robots are often underpowered and need to be frugal on resources for computing and reasoning - being able to move is their first priority!
展开▼