Failures in mobile robot navigation are often caused by errors in localizing the robot relative to its environment This paper explores the idea that these errors can be considerably reduced by planning paths taking the robot through positions where pertinent features of the environment can be sensed It introduces the notion of a "sensory uncertainty field" (SUF) For every possible robot configuration q, this field estimates the distribution of possible errors in the robot configuration that would be computed by a localization function matching the data given by the sensors against an environment model, if the robot was at q A planner is proposed which uses a precomputed SUF to generate paths that minimize expected errors or any other criterion combining, say, path length and errors This paper describes in detail the computation of a specific SUF for a mobile robot equipped with a classical line-striping camera/laser range sensor It presents an implemented SUF-based motion planner for this robot and shows paths generated by this planner Navigation experiments were conducted with mobile robots using paths generated by the SUF-based planner and other paths The former paths were tracked with greater precision than the others The final section of the paper discusses additional research issues related to SUF-based planning >

Planning the motions of a mobile robot in a sensory uncertainty field

Relevant research in human vision is summarized. Computational models for selective fixation control, most of which have begun to emerge only in recent years, are discussed. The methods examined involve several visual and nonvisual criteria, and the resulting sensing strategies are highly goal dependent. Both biological and machine vision are considered. >

A survey of selective fixation control for machine vision

This article describes planning of sensing and motion by a mobile robot to build the map of a 2D unknown environment. The map is represented as a graph that describes the structure of the skeleton of the environment. It is incrementally built by connecting the local structure of the skeleton extracted from sensory data at each position of sensing. The motion of the robot to observe currently unknown region is first roughly planned with the graph-based global map and then precisely planned with a geometric local sensory information.

Continuous purposive sensing and motion for 2D map building

Finding the global optimum of a multivariate function without testing every candidate solution involves making assumptions about the structure of the function's search space. Techniques that exploit only minimal assumptions often achieve their generality at the expense of practical efficiency. A theoretical analysis and two algorithms are presented for global optimization problems that involve articulated figures--jointed objects composed of rigid rods. In each case, it is found that assumptions specific to articulated figures are appropriate.
1. Computing the native fold of a protein by globally minimizing its potential-energy function is widely believed to require an amount of CPU time that is exponential in the number of soft degrees of freedom in the molecule. Conventional arguments as to this exponential time complexity are fallacious. A typical mathematical statement of the problem is shown to be NP-hard, and therefore intractable unless all NP-complete problems can be solved efficiently. This result suggests that if an algorithm is to be efficient for predicting protein structures, it must exploit assumptions about protein structures in addition to what can be inferred from the functional form of the potential energy. Furthermore, it is argued that the usual form of the Levinthal paradox, a central issue in protein folding, is not compelling because its premise requires the use of a physical model that contradicts existing experimental evidence. A reformulation of the paradox is presented, and the possible resolutions of the paradox are enumerated exhaustively.
2. The task of identifying all conformers of cycloheptadecane whose energies lie within 3 kcal/mol of its global optimum has been identified as a benchmark conformational search problem that is computationally challenging for contemporary algorithms and hardware. In principle, holonomic constraints on bond lengths and angles can be used to reduce the computational cost of the search, but in practice, exploiting all of these constraints simultaneously is far from straightforward because cycloheptadecane is a ring. An algorithm in which ring closure is maintained by kinematic techniques is presented. Compared to previous approaches, the new algorithm is faster and more thorough, and it is more amenable to analysis.
3. Animated characters are frequently modeled as articulated figures. Physically correct motions for articulated figures are difficult to generate. Current techniques for generating such motions are local in nature; they use perturbational analysis to refine an initial trajectory. Because of a multiple-minimum problem, this approach is inadequate for many applications. A stochastic global search technique is presented. The key elements of the algorithm are a method for encoding trajectories as behaviors, and a genetic algorithm for choosing behavior parameters that is implemented on a massively parallel computer.

Algorithms for automatic sensor placement to acquire complete and accurate information

We consider the view planning problem, also called, next best view (NBV) problem for sensor based exploration for general robot-sensor systems, where a range scanner is mounted on a robot with non-trivial kinematics, e.g., an eye-in-hand system. Earlier approaches to NBV had considered purely work-space (we also use the term physical-space) based criteria, such as select the view that maximizes the unknown physical-space volume. While this works well for mobile robots (often modeled as point or circle, thereby having trivial geometry and kinematics), it ignores a critical aspect, i.e., to give priority to exploring "manoeuvrable" space around the robot so that it can move to better viewing configurations. Proposed C-space based view planning criteria address this problem. However, C-space criteria (assuming the robot has enough manoeuvrable space) may sacrifice efficiency in work-space volume coverage. For inspection or environment modeling tasks, efficient workspace volume coverage is important. In this paper, we propose an adaptive algorithm that biases the search toward C-space or toward work-space, as needed. We call it adaptive viewpoint candidates entropy (VCE) criterion. Results with different simulated scenes show the effectiveness of this criterion in efficient (in that it needs less scans) exploration of the workspace.

An adaptive configuration-space and work-space based criterion for view planning

Two O(N log N) heuristic approaches for planning viewpoints in 2D known environments are described. Both are based on the generalized Delaunay triangulation of a polygon with holes. The main difference between them is the heuristic strategies used to merge the triangles to form the partitioning or covering star polygons. When the environment needs to be explored, an O(N/sup 2/ log N) heuristic scheme which successively selects the viewpoints and views for mobile robots is used. Upper bounds on the number of planned viewpoints and views for each scheme are estimated. >

S. Xie

Papers

View planning for mobile robots