Abstract
Localization has been regarded as one of the most fundamental problems to enable a mobile robot with autonomous capabilities. Probabilistic techniques such as Kalman or Particle filtering have long been used to solve robotic localization and mapping problem. Despite their good performance in practical applications, they could suffer inconsistency problems. This paper presents an Interval Constraint Satisfaction Problem (ICSP) graph based methodology for consistent car-like robot localization in outdoor environments. The localization problem is cast into a two-stage framework: visual teach and repeat. During a teaching phase, the interval map is built when a robot navigates around the environment with GPS-support. The map is then used for real-time ego-localization as the robot repeats the path autonomously. By dynamically solving the ICSP graph via Interval Constraint Propagation (ICP) techniques, a consistent and improved localization result is obtained. Both numerical simulation results and real data set experiments are presented, showing the soundness of the proposed method in achieving consistent localization.
Highlights
Localization is a problem for mobile robots to localize themselves in the environment with sensory information from their embedded sensors
The real-time localization result is obtained by using only the current knowledge of the observations; this localization result can be further improved by considering a dynamic Interval Constraint Satisfaction Problem (ICSP) graph in a sliding window during each timestep, such that the localization improvement due to the new observation can be propagated to the past through the graph architecture and improve the past localization
To test the feasibility of the proposed ICSP graph based visual teach method, we set up a simple numerical experiment
Summary
Localization is a problem for mobile robots to localize themselves in the environment with sensory information from their embedded sensors. While some others get rid of dealing with the uncertainty probability distributions, they assumed that noise is unknown but bounded by real intervals Those methods provide solutions presented by a set of bounded configurations in which the robot is guaranteed to be, and they are classified as deterministic methods. The main advantage of interval analysis based localization over Kalman filtering or Bayesian methods is that they provide guaranteed solutions without the need to linearize the robot motion or the sensoring models, unlike the probabilistic counterparts that require linearization to facilitate the propagation of uncertainties. Interval methods do not assume any noise probability distribution in the system; they just require a soft assumption about the support of the noise, i.e., it is bounded by real intervals They can provide guaranteed and consistent results.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
