Search Results

There are many well-documented peculiarities of human spatial cognition. Are these simply limitations of the human mind and brain, and the design for an intelligent robot should ignore or avoid them? Or are they unavoidable trade-offs in optimal strategies for solving the spatial problems faced by an autonomous agent? In our work on the Spatial Semantic Hierarchy (SSH), we take the latter position, focusing on the ability to express incomplete knowledge of space. At the Control level, the agent selects and follows control laws to define a discrete set of distinctive states within the continuous environment. These states, along with actions abstracting the control laws linking them, form the Causal level of the SSH. From this, we derive the Topological level, which describes the environment qualitatively in terms of places, paths, and regions, related by connectivity, order, and containment. The Hybrid SSH generalizes previous work by building local metrical maps directly from perceptual information, and by building global metrical maps on the skeleton provided by the global topological map. By having multiple representations for these kinds of incomplete spatial knowledge, we hope that the Spatial Semantic Hierarchy will contribute to understanding robustness and scalability of human spatial cognition. Papers and other information can be obtained at http://www.cs.utexas.edu/~qr/robotics/.