Active Exploration for Maximising Map AccuracyNicholas RoyWhatTremendous progress has been made recently in simultaneous localisation and mapping of unknown environments. Using sensor and odometry data from an exploring mobile robot, it has become much easier to build high-quality globally consistent maps of many large, real-world environments. However, relatively little attention has been given to the trajectories that generate the data for building these maps. Exploration strategies are often used to cover the largest amount of unknown space as quickly as possible, but few strategies exist for building the most reliable map possible. Nevertheless, the particular control strategy can have a substantial impact on the quality of the resulting map. We have developed control algorithms for exploring unknown space that explicitly tries to build as large a map as possible while maintaining as accurate a map as possible. WhyRecording the sensor data for a good map is not necessarily a straight-forward
process, even done manually. The control strategy used to acquire the
data can have a substantial impact on the quality of the resulting map;
different kinds of motions can lead to greater or smaller errors in the
mapping process. Ideally, we would like to automate the exploration process,
not just for efficient exploration of unknown areas of space, but exploration
that will lead to high accuracy maps. How We consider the problem of building
an accurate map in the Extended Kalman Filter framework [1].
The approach we take is to use a one-step lookahead, choosing trajectories
that maximise the space explored while minimising the likelihood we will
become lost on re-entering known space. In this case, the single step
is over a path from the existing map through unexplored space to the first
sensor reading from within the known map. At every time step, we will
choose a trajectory that will minimise our uncertainty as we re-enter
the map, at the same time maximising the coverage of the unexplored area.
We use a parameterised class of paths and repeatedly choose the parameter
setting that maximises our objective function. The class of paths is given
by the parametric curve [2]: which expresses the distance For a particular trajectory
This function contains the one free parameter Progress Given the policy class
FutureThe online strategy currently contains at some open problems. Firstly,
the reward function contains an explicit trade-off between exploration
and map accuracy, represented by the free parameter AcknowledgmentsThis is joint work with Robert Sim and Gregory Dudek. References[1]J. J. Leonard, I. J. Cox, and H. F. Durrant-Whyte. Dynamic map building for an autonomous mobile robot. Int. J. Robotics Research, 11(4):286--298, 1992. [2] Robert Sim and Gregory Dudek. Effective exploration strategies for the construction of visual maps. In Proc. IJCAI Workshop on Reasoning with Uncertainty in Robotics (RUR), pages 69--76, 2003. [3] Robert Sim, Gregory Dudek, and Nicholas Roy. Online control policy optimization for minimizing map uncertainty during exploration. In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA 2004), New Orleans, LA, April 2004. |
||||
|