Projects | Ranulfo Bezerra

Automation for Transformable Production

Wed, 27 Apr 2016 00:00:00 +0000

As the fashion industry shifts towards on-demand orders to accommodate rapidly evolving trends, Transformable Production emerges as a vital approach for personalized clothing manufacturing. Our project aims to optimize task allocation and path finding in a Transformable Production setting, utilizing both static and mobile robots for a more efficient and adaptable production process.

Disaster Prevention Challenge on World Robot Summit

Wed, 27 Apr 2016 00:00:00 +0000

In a noteworthy achievement, a multi-modal robot team excelled in the Disaster Prevention Challenge during the World Robot Summit 2020. This post delves into the innovative system developed to automate industrial plant inspections, comprised of a tracked vehicle, a mecanum-wheeled vehicle, and an unmanned aerial vehicle (UAV). Two operators efficiently manage the team, focusing on work-related and visual inspection tasks. The post will highlight experimental results, showcasing each robot type’s specialized capabilities: the tracked vehicle excels in challenging mechanical tasks, the mecanum-wheeled vehicle performs rapid visual inspections in narrow passages, and the UAV offers quick aerial assessments. The team’s remarkable performance metrics in the competition’s final round will also be discussed, including high completion rates in work-related tasks (57%), visual inspections (85%), and emergency responses (92%).

Knowledge Acquisition from Sparse Mobile Probe Data

Wed, 27 Apr 2016 00:00:00 +0000

The “Knowledge Acquisition from Sparse Mobile Probe Data” project addresses challenges faced by autonomous driving systems (ADSs) in urban environments through three interconnected tasks. The first task improves pedestrian trajectory analysis for safer navigation. The second task identifies causes of vehicle behavior changes in different areas, providing insights for more adaptive and responsive ADSs. The third task recognizes traffic regions using driving log data, enhancing navigation capabilities even without visual information. By integrating these techniques, ADSs gain a comprehensive understanding of their environment, leading to better decision-making and safer navigation in complex urban settings. This project aims to overcome current limitations in autonomous driving technology and build public trust.

Optimizing Construction Vehicle Interaction with Machine Learning

Wed, 27 Apr 2016 00:00:00 +0000

In the realm of earthmoving work, the interaction between autonomous dump trucks and human-operated backhoes is critical for efficient and safe operations. Recent studies have focused on leveraging machine learning techniques, particularly the Beta-Process Hidden Markov Model (BP-HMM), to intelligently predict backhoe loading times. One study developed a BP-HMM-based prediction method that automatically identifies the transition of several primitive motions from time-series data, achieving a remarkable 100% accuracy rate. This algorithm enhances the robustness of prediction models across different operator-backhoe combinations and sensor layouts. Another study used 6-axis IMU sensors to collect time-series data from backhoes, and employed BP-HMM to model specific operator behaviors. The model was able to predict the loading instant with up to 100% probability, significantly reducing idle time and risk for dump trucks. These advancements contribute significantly to the automation and safety of construction vehicles, enabling seamless cooperation between autonomous and human-operated machinery.