CIR, CIS, & Mines: Triple AI, Robotics and Computation Event
Room: 222, Bldg: Marquez Hall, 1600 Arapahoe St, Golden, Colorado, United States, 80401November 14, 2024, 4:00 PM – 6:00 PM (MDT) AI, Robotics, and Computation Presentations: Multi-modal communication in Multi-task context Global Point Cloud Registration in Structured Scenes Navigating the Energy Frontier: Efficient Motion Planning for Mobile Robots Speaker(s): Yifei (Rena) Zhu, Abolfazl Babanazari, Tanmay Desai Agenda: November 14, 2024, 4:00 PM – 6:00 PM (MDT) Yifei (Rena) Zhu Colorado School of Mines PhD Student in Robotics Rena is a third-year PhD student in Robotics in the Mines Interactive Robotics Research Lab at the Colorado School of Mines. Rena is interested in exploring human-centered design paradigms within Mixed/Augmented Reality and Human-robot interaction (HRI), and understanding what role MR/AR technology can, and should, play in HRI through an interdisciplinary lens. Moreover, Rena is interested in technology ethics, AI ethics, and technology policy, and wants to explore how research efforts may impact society at large. Presentation: Understanding Multi-modal communication in Multi-task context Abstract: Multiple Resource Theory suggests that the mind has largely distinct types of cognitive resources associated with different sensory modalities, with subdivisions along a number of other dimensions. In this work we explore how the modality of communication used by task guidance systems meant to assist in multi-task contexts interacts with the modality of underlying tasks to determine overall cognitive load and task performance. Our results suggest that users can strategically avoid overload by selectively attending to dimensions of multi-modal communication. Abolfazl Babanazari Colorado School of Mines PhD in Computer Science Abolfazl is a graduate student currently pursuing a PhD in Computer Science at the Colorado School of Mines under supervision of Kaveh Fathian. Abolfazl has prior experience working in the medical field on stereo systems and developing AR systems for surgical navigation. Abolfazl's research interests include Linear Algebra, Graph Theory, and Optimization, particularly in their applications to Robotics and Autonomy, such as Simultaneous Localization and Mapping (SLAM) and perception. Abolfazl is currently focusing on robust data association techniques from both a theoretical and practical aspect. Presentation: Global Point Cloud Registration in Structured Scenes Abstract: Point cloud registration, especially without an initial transformation, is crucial for many tasks in robotics and computer vision. When the corresponding parts of two point clouds are unknown, the solution search space grows exponentially with the size of the point clouds. A widely used approach to address this issue involves leveraging local surface properties around given points and matching these to corresponding points across point clouds. While effective, this technique struggles in environments with repetitive patterns, such as man-made structures, urban areas, or indoor settings, where point-based matching methods can generate associations with an extremely high outlier ratio. In this presentation, we explore the potential of using primitive mathematical shapes and the possibility of integrating them into existing Global Point-Based Registration (GPBR) algorithms. By reducing the number of associations and outlier ratio, this technique can significantly improve the accuracy and robustness of point cloud registration. Tanmay Desai Colorado School of Mines PhD student in Robotics Tanmay is a second-year PhD student in Robotics advised by Dr. Iris Bahar at the Colorado School of Mines. Tanmay is interested in exploring better motion planning algorithms and how can we utilize hardware accelerators specifically Field Programmable Gate Arrays (FPGAs) to get power efficient and faster convergence to controls in stochastic environments. Presentation: Navigating the Energy Frontier: Efficient Motion Planning for Mobile Robots Abstract: Mobile robots often struggle with the computational demands of sampling-based motion planning algorithms. This research explores the potential of FPGAs as a more energy-efficient alternative to GPUs. By leveraging their parallel architecture and hardware-specific optimizations, FPGAs could significantly improve the performance-per-watt ratio of motion planning systems. Room: 222, Bldg: Marquez Hall, 1600 Arapahoe St, Golden, Colorado, United States, 80401