Self-adaptive Monte Carlo localization for mobile robots using range finders

Adaptive Model Predictive Control for Mobile Robots with Localization Fluctuation Estimation

Mobile robots are widely employed in various fields to perform autonomous tasks. In dynamic scenarios, localization fluctuations are unavoidable and obvious. However, common controllers do not consider the impact of localization fluctuations, resulting in violent jittering or poor trajectory tracking of the mobile robot. For this reason, this paper proposes an adaptive model predictive control (MPC) with an accurate localization fluctuation assessment for mobile robots, which balances the contradiction between precision and calculation efficiency of mobile robot control. The distinctive features of the proposed MPC are three-fold: (1) Integrating variance and entropy-a localization fluctuation estimation relying on fuzzy logic rules is proposed to enhance the accuracy of the fluctuation assessment. (2) By using the Taylor expansion-based linearization method-a modified kinematics model that considers that the external disturbance of localization fluctuation is established to satisfy the iterative solution of the MPC method and reduce the computational burden. (3) An improved MPC with an adaptive adjustment of predictive step size according to localization fluctuation is proposed, which alleviates the disadvantage of a large amount of the MPC calculation and improves the stability of the control system in dynamic scenes. Finally, verification experiments of the real-life mobile robot are offered to verify the effectiveness of the presented MPC method. Additionally, compared with PID, the tracking distance and angle error of the proposed method decrease by 74.3% and 95.3%, respectively.

Vision-Sensor-Assisted Probabilistic Localization Method for Indoor Environment

Among the numerous indoor localization methods, Light-Detection-and-Ranging (LiDAR)-based probabilistic algorithms have been extensively applied to indoor localization due to their real-time performance and high accuracy. Nevertheless, these methods are challenged in symmetrical environments when tackling global localization and the robot kidnapping problem. In this paper, a novel hybrid method that combines visual and probabilistic localization results is proposed. Augmented Monte Carlo Localization (AMCL) is improved for position tracking continually. LiDAR-based measurements' uncertainty is evaluated to incorporate discrete visual-based results; therefore, a better diversity of the particle can be maintained. The robot kidnapping problem can be detected and solved by preventing premature convergence of the particle filter. Extensive experiments were implemented to validate the robustness and accuracy performance. Meanwhile, the localization error was reduced from 30 mm to 9 mm during a 600 m tour.

Velocity Estimation and Cost Map Generation for Dynamic Obstacle Avoidance of ROS Based AMR

In the past few years, due to the growth of the open-source community and the popularity of perceptual computing resources, the ROS (Robotic Operating System)Ecosystem has been widely shared and used in academia, industrial applications, and service fields. With the advantages of reusability of algorithms and system modularity, service robot applications are flourishing via the released ROS navigation framework. In the ROS navigation framework, the grid cost maps are majorly designed for path planning and obstacle avoidance with range sensors. However, the robot will often collide with dynamic obstacles since the velocity information is not considered within the navigation framework in time. This study aims to improve the feasibility of high-speed dynamic obstacle avoidance for an ROS-based mobile robot. In order to enable the robot to detect and estimate dynamic obstacles from a first-person perspective, vision tracking and a laser ranger with an Extend Kalman Filter (EKF) have been applied. In addition, an innovative velocity obstacle layer with truncated distance is implemented for the path planner to analyze the performances between the simulated and actual avoidance behavior. Finally, via the velocity obstacle layer, as the robot faces the high-speed obstacle, safe navigation can be achieved.

Efficient Detection of Robot Kidnapping in Range Finder-Based Indoor Localization Using Quasi-Standardized 2D Dynamic Time Warping

This paper proposes an augmented online approach to detect kidnapping events within range-finder-based indoor localization. The method is specifically designed for an Internet of Things (IoT)-Aided Robotics Platform that enables the system to detect kidnapping across all time instances of an indoor mobile robotic operation with high accuracy and maintain a high accuracy in the face of relocalization failures. The approach is based on similarity degree of geometry shape of the environment obtained from range scan data between two consecutive time instances. The proposed approach named Quasi-Standardized Two-Dimensional Dynamic Time Warping (QS-2DDTW) is based on the Multidimensional Dynamic Time Warping (MD-DTW) with homogeneity variance test imbued in it. A series of simulations are preformed against maximum current weight, measurement entropy, and the four metrics in metric based detector. The result shows that the proposed method yields high performance in terms of its ability to distinguish kidnapping condition from normal condition and that it has low dependency towards relocalization process, thus ensures the accuracy of detection is not disturbed by relocalization.

Deep Reinforcement Learning for Indoor Mobile Robot Path Planning

This paper proposes a novel incremental training mode to address the problem of Deep Reinforcement Learning (DRL) based path planning for a mobile robot. Firstly, we evaluate the related graphic search algorithms and Reinforcement Learning (RL) algorithms in a lightweight 2D environment. Then, we design the algorithm based on DRL, including observation states, reward function, network structure as well as parameters optimization, in a 2D environment to circumvent the time-consuming works for a 3D environment. We transfer the designed algorithm to a simple 3D environment for retraining to obtain the converged network parameters, including the weights and biases of deep neural network (DNN), etc. Using these parameters as initial values, we continue to train the model in a complex 3D environment. To improve the generalization of the model in different scenes, we propose to combine the DRL algorithm Twin Delayed Deep Deterministic policy gradients (TD3) with the traditional global path planning algorithm Probabilistic Roadmap (PRM) as a novel path planner (PRM+TD3). Experimental results show that the incremental training mode can notably improve the development efficiency. Moreover, the PRM+TD3 path planner can effectively improve the generalization of the model.

Reliable and Fast Localization in Ambiguous Environments Using Ambiguity Grid Map

In real-world robotic navigation, some ambiguous environments contain symmetrical or featureless areas that may cause the perceptual aliasing of external sensors. As a result of that, the uncorrected localization errors will accumulate during the localization process, which imposes difficulties to locate a robot in such a situation. Using the ambiguity grid map (AGM), we address this problem by proposing a novel probabilistic localization method, referred to as AGM-based adaptive Monte Carlo localization. AGM has the capacity of evaluating the environmental ambiguity with average ambiguity error and estimating the possible localization error at a given pose. Benefiting from the constructed AGM, our localization method is derived from an improved Dynamic Bayes network to reason about the robot's pose as well as the accumulated localization error. Moreover, a portal motion model is presented to achieve more reliable pose prediction without time-consuming implementation, and thus the accumulated localization error can be corrected immediately when the robot moving through an ambiguous area. Simulation and real-world experiments demonstrate that the proposed method improves localization reliability while maintains efficiency in ambiguous environments.

A Robust Indoor Localization System Integrating Visual Localization Aided by CNN-Based Image Retrieval with Monte Carlo Localization

This paper proposes a novel multi-sensor-based indoor global localization system integrating visual localization aided by CNN-based image retrieval with a probabilistic localization approach. The global localization system consists of three parts: coarse place recognition, fine localization and re-localization from kidnapping. Coarse place recognition exploits a monocular camera to realize the initial localization based on image retrieval, in which off-the-shelf features extracted from a pre-trained Convolutional Neural Network (CNN) are adopted to determine the candidate locations of the robot. In the fine localization, a laser range finder is equipped to estimate the accurate pose of a mobile robot by means of an adaptive Monte Carlo localization, in which the candidate locations obtained by image retrieval are considered as seeds for initial random sampling. Additionally, to address the problem of robot kidnapping, we present a closed-loop localization mechanism to monitor the state of the robot in real time and make adaptive adjustments when the robot is kidnapped. The closed-loop mechanism effectively exploits the correlation of image sequences to realize the re-localization based on Long-Short Term Memory (LSTM) network. Extensive experiments were conducted and the results indicate that the proposed method not only exhibits great improvement on accuracy and speed, but also can recover from localization failures compared to two conventional localization methods.

A Q-learning approach based on human reasoning for navigation in a dynamic environment

A Q-learning approach is often used for navigation in static environments where state space is easy to define. In this paper, a new Q-learning approach is proposed for navigation in dynamic environments by imitating human reasoning. As a model-free method, a Q-learning method does not require the environmental model in advance. The state space and the reward function in the proposed approach are defined according to human perception and evaluation, respectively. Specifically, approximate regions instead of accurate measurements are used to define states. Moreover, due to the limitation of robot dynamics, actions for each state are calculated by introducing a dynamic window that takes robot dynamics into account. The conducted tests show that the obstacle avoidance rate of the proposed approach can reach 90.5% after training, and the robot can always operate below the dynamics limitation.

Design of a Hybrid Indoor Location System Based on Multi-Sensor Fusion for Robot Navigation

In the case of a single scene feature, the positioning of an indoor service robot takes a long time, and localization errors are likely to occur. A new method for a hybrid indoor localization system according to multi-sensor fusion is proposed to solve these problems. The localization process is divided in two stages: rough positioning and precise positioning. By virtue of the K nearest neighbors based on possibility (KNNBP) algorithm first created in the present study, the rough position of a robot is determined according to the received signal strength indicator (RSSI) of Wi-Fi. Then, the hybrid particle filter localization (HPFL) algorithm improved on the basis of adaptive Monte Carlo localization (AMCL) is adopted to get the precise localization, which integrates various information, including the rough position and information from Lidar, a compass, an occupancy grid map, and encoders. The experiments indicated that the positioning error was 0.05 m; the success rate of localization was 96% with even 3000 particles, and the global positioning time was 1.9 s. However, under the same conditions, the success rate of AMCL was approximately 40%, the required time was approximately 25.6 s, and the positioning accuracy was the same. This indicates that the hybrid indoor location system is efficient and accurate.

Self-Diagnosis of Localization Status for Autonomous Mobile Robots

It is essential to provide reliable localization results to allow mobile robots to navigate autonomously. Even though many state-of-the-art localization schemes have so far shown satisfactory performance in various environments, localization has still been difficult under specific conditions, such as extreme environmental changes. Since many robots cannot diagnose for themselves whether the localization results are reliable, there can be serious autonomous navigation problems. To solve this problem, this study proposes a self-diagnosis scheme for the localization status. In this study, two indicators are empirically defined for the self-diagnosis of localization status. Each indicator shows significant changes when there are difficulties in light detection and ranging (LiDAR) sensor-based localization. In addition, the classification model of localization status is trained through machine learning using the two indicators. A robot can diagnose the localization status itself using the proposed classification model. To verify the usefulness of the proposed method, we carried out localization experiments in real environments. The proposed classification model successfully detected situations where the localization accuracy is significantly degraded due to extreme environmental changes.

Detection of kidnapped robot problem in Monte Carlo localization based on the natural displacement of the robot

This article proposes a new method to detect the kidnapped robot problem event in Monte Carlo localization. The method is designed in such a manner that it can provide accurate detection across all time instances, whether the robot can still recognize part of the environment or is totally lost after kidnapping. The proposed method uses the sensor reading of the robot to determine if robot’s displacement at particular time instance is considered a natural displacement or not. A series of simulations are designed to measure the accuracy of detection and how it compares to other methods. The simulations show that the proposed method outperforms the methods of detection based on the weight of particles.

Enhanced Monte Carlo localization incorporating a mechanism for preventing premature convergence

In this paper, we propose an enhanced Monte Carlo localization (EMCL) algorithm for mobile robots, which deals with the premature convergence problem in global localization as well as the estimation error existing in pose tracking. By incorporating a mechanism for preventing premature convergence (MPPC), which uses a “reference relative vector” to modify the weight of each sample, exploration of a highly symmetrical environment can be improved. As a consequence, the proposed method has the ability to converge particles toward the global optimum, resulting in successful global localization. Furthermore, by applying the unscented Kalman Filter (UKF) to the prediction state and the previous state of particles in Monte Carlo Localization (MCL), an EMCL can be established for pose tracking, where the prediction state is modified by the Kalman gain derived from the modified prior error covariance. Hence, a better approximation that reduces the discrepancy between the state of the robot and the estimation can be obtained. Simulations and practical experiments confirmed that the proposed approach can improve the localization performance in both global localization and pose tracking.

Kullback-Leibler Divergence-Based Differential Evolution Markov Chain Filter for Global Localization of Mobile Robots

One of the most important skills desired for a mobile robot is the ability to obtain its own location even in challenging environments. The information provided by the sensing system is used here to solve the global localization problem. In our previous work, we designed different algorithms founded on evolutionary strategies in order to solve the aforementioned task. The latest developments are presented in this paper. The engine of the localization module is a combination of the Markov chain Monte Carlo sampling technique and the Differential Evolution method, which results in a particle filter based on the minimization of a fitness function. The robot’s pose is estimated from a set of possible locations weighted by a cost value. The measurements of the perceptive sensors are used together with the predicted ones in a known map to define a cost function to optimize. Although most localization methods rely on quadratic fitness functions, the sensed information is processed asymmetrically in this filter. The Kullback-Leibler divergence is the basis of a cost function that makes it possible to deal with different types of occlusions. The algorithm performance has been checked in a real map. The results are excellent in environments with dynamic and unmodeled obstacles, a fact that causes occlusions in the sensing area.

Active global localization based on localizability for mobile robots

In global localization under the framework of a particle filter, the acquiring of effective observations of the whole particle system will be greatly effected by the uncertainty of a prior-map, such as unspecific structures and noises. In this study, taking the uncertainty of the prior-map into account, a localizability-based action selection mechanism for mobile robots is proposed to accelerate the convergence of global localization. Localizability is defined to evaluate the observations according to the prior-map (probabilistic grid map) and observation (laser range-finder) models based on the Cramér-Rao Bound. The evaluation considers the uncertainty of the prior-map and does not need to extract any specific observation features. Essentially, localizability is the determinant of the inverse covariance matrix for localization. Specifically, at the beginning of every filtering step, the action, which makes the whole particle system to achieve the maximum localizability distinctness, is selected as the actual action. Then there will be the increased opportunities for accelerating the convergence of the particles, especially in the face of the prior-map with uncertainty. Additionally, the computational complexity of the proposed algorithm does not increase significantly, as the localizability is pre-cached off-line. In simulations, the proposed active algorithm is compared with the passive algorithm (i.e. global localization with the random robot actions) in environments with different degrees of uncertainty. In experiments, the effectiveness of the localizability is verified and then the comparative experiments are conducted based on an intelligent wheelchair platform in a real environment. Finally, the experimental results are compared and analyzed among the existing active algorithms. The results demonstrate that the proposed algorithm could accelerate the convergence of global localization and enhance the robustness against the system ambiguities, thereby reducing the failure probability of the convergence.