The Deep Convolutional Neural Network Role in the Autonomous Navigation of Mobile Robots (SROBO)

A new CNN-BASED object detection system for autonomous mobile robots based on real-world vehicle datasets

Recently, autonomous mobile robots (AMRs) have begun to be used in the delivery of goods, but one of the biggest challenges faced in this field is the navigation system that guides a robot to its destination. The navigation system must be able to identify objects in the robot's path and take evasive actions to avoid them. Developing an object detection system for an AMR requires a deep learning model that is able to achieve a high level of accuracy, with fast inference times, and a model with a compact size that can be run on embedded control systems. Consequently, object recognition requires a convolutional neural network (CNN)-based model that can yield high object classification accuracy and process data quickly. This paper introduces a new CNN-based object detection system for an AMR that employs real-world vehicle datasets. First, we create original real-world datasets of images from Banda Aceh city. We then develop a new CNN-based object identification system that is capable of identifying cars, motorcycles, people, and rickshaws under morning, afternoon, and evening lighting conditions. An SSD Mobilenetv2 FPN Lite 320 × 320 architecture is employed for retraining using these real-world datasets. Quantitative and qualitative performance indicators are then applied to evaluate the CNN model. Training the pre-trained SSD Mobilenetv2 FPN Lite 320 × 320 model improves its classification and detection accuracy, as indicated by its performance results. We conclude that the proposed CNN-based object detection system has the potential for use in an AMR.

A Performance Comparison between Different Industrial Real-Time Indoor Localization Systems for Mobile Platforms

The flexibility and versatility associated with autonomous mobile robots (AMR) have facilitated their integration into different types of industries and tasks. However, as the main objective of their implementation on the factory floor is to optimize processes and, consequently, the time associated with them, it is necessary to take into account the environment and congestion to which they are subjected. Localization, on the shop floor and in real time, is an important requirement to optimize the AMRs' trajectory management, thus avoiding livelocks and deadlocks during their movements in partnership with manual forklift operators and logistic trains. Threeof the most commonly used localization techniques in indoor environments (time of flight, angle of arrival, and time difference of arrival), as well as two of the most commonly used indoor localization methods in the industry (ultra-wideband, and ultrasound), are presented and compared in this paper. Furthermore, it identifies and compares three industrial indoor localization solutions: Qorvo, Eliko Kio, and Marvelmind, implemented in an industrial mobile platform, which is the main contribution of this paper. These solutions can be applied to both AMRs and other mobile platforms, such as forklifts and logistic trains. In terms of results, the Marvelmind system, which uses an ultrasound method, was the best solution.

Deep Bayesian-Assisted Keypoint Detection for Pose Estimation in Assembly Automation

Pose estimation is crucial for automating assembly tasks, yet achieving sufficient accuracy for assembly automation remains challenging and part-specific. This paper presents a novel, streamlined approach to pose estimation that facilitates automation of assembly tasks. Our proposed method employs deep learning on a limited number of annotated images to identify a set of keypoints on the parts of interest. To compensate for network shortcomings and enhance accuracy we incorporated a Bayesian updating stage that leverages our detailed knowledge of the assembly part design. This Bayesian updating step refines the network output, significantly improving pose estimation accuracy. For this purpose, we utilized a subset of network-generated keypoint positions with higher quality as measurements, while for the remaining keypoints, the network outputs only serve as priors. The geometry data aid in constructing likelihood functions, which in turn result in enhanced posterior distributions of keypoint pixel positions. We then employed the maximum a posteriori (MAP) estimates of keypoint locations to obtain a final pose, allowing for an update to the nominal assembly trajectory. We evaluated our method on a 14-point snap-fit dash trim assembly for a Ford Mustang dashboard, demonstrating promising results. Our approach does not require tailoring to new applications, nor does it rely on extensive machine learning expertise or large amounts of training data. This makes our method a scalable and adaptable solution for the production floors.

A Machine Learning Approach to Robot Localization Using Fiducial Markers in RobotAtFactory 4.0 Competition

Localization is a crucial skill in mobile robotics because the robot needs to make reasonable navigation decisions to complete its mission. Many approaches exist to implement localization, but artificial intelligence can be an interesting alternative to traditional localization techniques based on model calculations. This work proposes a machine learning approach to solve the localization problem in the RobotAtFactory 4.0 competition. The idea is to obtain the relative pose of an onboard camera with respect to fiducial markers (ArUcos) and then estimate the robot pose with machine learning. The approaches were validated in a simulation. Several algorithms were tested, and the best results were obtained by using Random Forest Regressor, with an error on the millimeter scale. The proposed solution presents results as high as the analytical approach for solving the localization problem in the RobotAtFactory 4.0 scenario, with the advantage of not requiring explicit knowledge of the exact positions of the fiducial markers, as in the analytical approach.

Posture and Map Restoration in SLAM Using Trajectory Information

SLAM algorithms generally use the last system posture to estimate its current posture. Errors in the previous estimations can build up and cause significant drift accumulation. This accumulation of error leads to the bias of choosing accuracy over robustness. On the contrary, sensors like GPS do not accumulate errors. But the noise distribution in the readings makes it difficult to apply in high-frequency SLAM systems. This paper presents an approach which uses the advantage of both tightly-coupled SLAM systems and highly robust absolute positioning systems to improve the robustness and accuracy of a SLAM process. The proposed method uses a spare reference trajectory frame to measure the trajectory of the targeted robotic system and use it to recover the system posture during the mapping process. This helps the robotic system to reduce its accumulated error and able the system to recover from major mapping failures. While the correction process happens whenever a gap is detected between the two trajectories, the external frame does not have to be always available. The correction process is only triggered when the spare trajectory sensors can communicate. Thus, it reduces the needed computational power and complexity. To further evaluate the proposed method, the algorithm was assessed in two field tests and a public dataset. We have demonstrated that the proposed algorithm has the ability to be adapted into different SLAM approaches with various map representations. To share our findings, the software constructed for this project is open-sourced on Github.