Naoki Akai

Gaussian processes for magnetic map-based localization in large-scale indoor environments

The magnetic field that exists in an indoor environment includes rich magnetic fluctuations because buildings contain many magnetized materials (e.g., steel frames). These fluctuations can be used as landmarks, the use of which requires the creation of a magnetic map representing the distribution of the magnetic field. It is, however, difficult to build a large-scale magnetic map because of the narrow measurement range of a magnetic sensor. This paper proposes an efficient method for collecting magnetic data using a mobile robot and a method for building a magnetic map using Gaussian processes. The use of these methods make it possible to build a large-scale magnetic map efficiently. Moreover, this paper presents a particle filter-based localization method based on the magnetic map. The presented system enables a robot to identify its own position in large-scale buildings. Experiments are used to demonstrate the performance and usefulness of the presented system.

Monte Carlo Localization using magnetic sensor and LIDAR for real world navigation

For realizing more stable outdoor navigation for mobile robots, this paper proposes a localization method using a magnetic sensor and a Light Detection and Ranging (LIDAR). In the proposed method, Monte Carlo Localization (MCL) using the LIDAR and a determination method of a heading direction using the ambient magnetic field are combined. In other words, the proposal distribution becomes dense at the true state of the robot by using the ambient magnetic field. The determination method is based on the advantage of the magnetic navigation proposed by us. By the proposed method, the robot enabled to navigate with accuracy in the outdoor environment, since the robust localization is realized. The effectiveness of the proposed method is shown through experiments. Moreover, two robots implemented the proposed method achieved the task of Real World Robot Challenge 2012. This means that the proposed method is effective for real world navigation.

Robust localization using 3D NDT scan matching with experimentally determined uncertainty and road marker matching

In this paper, we present a localization approach that is based on a point-cloud matching method (normal distribution transform “NDT”) and road-marker matching based on the light detection and ranging intensity. Point-cloud map-based localization methods enable autonomous vehicles to accurately estimate their own positions. However, accurate localization and “matching error” estimations cannot be performed when the appearance of the environment changes, and this is common in rural environments. To cope with these inaccuracies, in this work, we propose to estimate the error of NDT scan matching beforehand (off-line). Then, as the vehicle navigates in the environment, the appropriate uncertainty is assigned to the scan matching. 3D NDT scan matching utilizes the uncertainty information that is estimated off-line, and is combined with a road-marker matching approach using a particle-filtering algorithm. As a result, accurate localization can be performed in areas in which 3D NDT failed. In addition, the uncertainty of the localization is reduced. Experimental results show the performance of the proposed method.

A navigation method based on topological magnetic and geometric maps for outdoor mobile robots

This paper proposes a novel navigation method based on topological magnetic and geometric maps for outdoor mobile robots. In this method, a state of a robot is represented by a travel distance and a lateral error, which is a distance from the travel path to the robot. In order to estimate the travel distance, a magnetic map which represents a magnetic field of the travel path is used. Since the magnetic field does not depend on moving objects, the travel distance can be robustly estimated against changes in geometric situations. However, deviating from the travel path is a fatal problem since the magnetic map does not record other magnetic fields of the travel path. To compensate this deviation, the lateral error is estimated by using geometric landmarks. Since this estimation is performed after the travel distance estimate, the lateral error can be stably estimated even if many moving objects surround the robot. As a result, the robot exactly navigates the travel path. Simulation and actual experiments are conducted to show the effectiveness and performance of the proposed method.

Pure Pursuit Revisited: Field Testing of Autonomous Vehicles in Urban Areas

In this paper, we aim to explore path following. We implement a path following component by referring to the existing Pure Pursuit algorithm. Using the simulation and field operational test, we identified the problem in the path following component. The main problems identified were with respect to vehicles meandering off the path, turning a corner, and the instability of steering control. Therefore, we apply some modifications to the Pure Pursuit[1] algorithm. We have also conducted the simulation and field operational tests again to evaluate these modifications.

Precise color extraction method based on color transition due to change in exposure time

This paper proposes a novel color extraction method using multiple images taken with different exposure time. Although using the multiple images is a famous technique to recover a high dynamic range radiance map, we do not use it. Alternatively, we focus on that any color performance is uniformly transited by changing exposure time even if under various illumination conditions. In this paper, we indicate it experimentally and develop a color extraction method with this characteristic. In the experiment, we set orange color as desired color and conducted color extraction to orange, red, and yellow colors in outdoor environments. Our proposed method exactly extracted only orange color from images under extreme exposure conditions.

Simultaneous pose and reliability estimation using convolutional neural network and Rao–Blackwellized particle filter

ABSTRACT In this study, we propose a novel localization approach that simultaneously estimates the reliability of estimation results. In the approach, a convolutional neural network (CNN) is used to make decision whether the localization process has failed or not. We train the CNN using a dataset that includes successful localization results and faults. However, the decision will contain some noise and many misdetection results may occur when the decision made by the CNN is used directly to detect faults. Therefore, we estimate both a robots pose and reliability of the localization results based on the decision. To simultaneously estimate the robots pose and reliability, we propose a new graphical model and implement a Rao–Blackwellized particle filter based on the model. We evaluated the proposed approach based on simulations and actual environments, which showed that the reliability estimated by the proposed approach can be used as an exact criterion for detecting localization faults. In addition, we show that the proposed approach can be applied in actual environments even when a dataset created from a simulation is used to train the CNN. GRAPHICAL ABSTRACT

Localization based on multiple visual-metric maps

This paper presents a fusion of monocular camera-based metric localization, IMU and odometry in dynamic environments of public roads. We build multiple vision-based maps and use them at the same time in localization phase. For the mapping phase, visual maps are built by employing ORB-SLAM and accurate metric positioning from LiDAR-based NDT scan matching. This external positioning is utilized to correct for scale drift inherent in all vision-based SLAM methods. Next in the localization phase, these embedded positions are used to estimate the vehicle pose in metric global coordinates using solely monocular camera. Furthermore, to increase system robustness we also proposed utilization of multiple maps and sensor fusion with odometry and IMU using particle filter method. Experimental testing were performed through public road environment as far as 170 km at different times of day to evaluate and compare localization results of vision-only, GNSS and sensor fusion methods. The results show that sensor fusion method offers lower average errors than GNSS and better coverage than vision-only one.

3D magnetic field mapping in large-scale indoor environment using measurement robot and Gaussian processes

Magnetic fields are used for localization and navigation in the field of robotics. In recent years, because of the spread of mobile devices equipped with magnetic sensors (e.g., smart phones), the use of magnetic fields has been extensive, especially for position tracking of mobile devices. One example application of such tracking is in identifying the position of a person with a mobile device. Development of this application requires a three-dimensional (3D) magnetic map that represents the magnetic distribution of a 3D environment since the device moves around in 3D space. It is, however, difficult to construct a 3D magnetic map of a large-scale environment because measuring the magnetic field is time consuming and expensive. In this paper we propose an efficient method for mapping the 3D magnetic field of a large-scale environment. The method uses a mobile manipulator to measure the 3D magnetic field, enabling 3D magnetic data to be automatically collected. Moreover, the method uses Gaussian processes (GPs) for regression of the magnetic field. In this study, we first evaluate the performance of the GPs and then describe the measurement robot. In an experiment, a 3D magnetic field of an indoor environment is visualized by using this method and the performance of the presented method is demonstrated.

Proactive driving modeling in blind intersections based on expert driver data

This paper presents a model for velocity control in blind corners and intersections based on expert driver data. Accurate expert driver data was collected with a car equipped with a 3D LiDAR and high definition maps. A model based on human expert driver data is used to control the velocity of the ego-vehicle when facing blind intersections. The model regulates ego-vehicle velocity based on the visibility of the road at the blind intersection. As the vehicle approximates the intersection and crossing roads are not visible, the vehicle slows down, then as the roads become visible the vehicle accelerates. Experimental results show the performance of the velocity model compared towards 270 trajectories taken from 7 expert drivers towards 6 different intersections without mandatory stops.