CU-Multi: A Dataset for Multi-Robot Data Association

Abstract: Multi-robot systems (MRSs) are valuable for tasks such as search and rescue due to their ability to coordinate over shared observations. A central challenge in these systems is aligning independently collected perception data across space and time, i.e., multi-robot data association. While recent advances in collaborative SLAM (C-SLAM), map merging, and inter-robot loop closure detection have significantly progressed the field, evaluation strategies still predominantly rely on splitting a single trajectory from single-robot SLAM datasets into multiple segments to simulate multiple robots. Without careful consideration to how a single trajectory is split, this approach will fail to capture realistic pose-dependent variation in observations of a scene inherent to multi-robot systems. To address this gap, we present CU-Multi, a multi-robot dataset collected over multiple days at two locations on the University of Colorado Boulder campus. Using a single robotic platform, we generate four synchronized runs with aligned start times and deliberate percentages of trajectory overlap. CU-Multi includes RGB-D, GPS with accurate geospatial heading, and semantically annotated LiDAR data. By introducing controlled variations in trajectory overlap and dense lidar annotations, CU-Multi offers a compelling alternative for evaluating methods in multi-robot data association. Instructions on accessing the dataset, support code, and the latest updates are publicly available at https://arpg.github.io/cumulti

Overview of CU-Multi: A Dataset for Multi-Robot Data Association

The paper presents the CU-Multi dataset, specifically designed to address the challenges associated with multi-robot data association in autonomous systems. These systems have substantial applications, particularly in scenarios requiring search and rescue operations, due to their integral ability to coordinate multi-faceted observations across spatial and temporal domains. A central issue within multi-robot systems (MRSs) is the alignment and integration of independently gathered perception data, often complicated by simultaneous spatial-temporal misalignments.

The CU-Multi dataset fundamentally attempts to bridge gaps that exist within current evaluation methodologies primarily composed of segmented single-robot SLAM datasets. Previous strategies that leverage single trajectory SLAM datasets divided into segments inadequately represent the realistic pose-dependent observation variance intrinsic to actual multi-robot operations. CU-Multi, hosted by the University of Colorado Boulder campus, provides a robust, realistic setting for evaluating the efficacy of collaborative SLAM (C-SLAM) algorithms, map merging techniques, and inter-robot loop closure detections by integrating accurately synchronized runs with initial trajectory alignment start times and controlled overlap percentages.

The dataset is composed of RGB-D, geospatially accurate GPS coordinates, and semantically annotated LiDAR data. These components act in concert to address critical aspects within multi-robot data association, where extensive trajectory overlap simulation is vital for advancing methods fostering MRSs.

Key Features and Dataset Specifications

The CU-Multi dataset comprises several notable features conducive for the meticulous evaluation of multi-robot data association methodologies:

• Synchronized Robotic Runs: The dataset includes four synchronized robotic runs collected over multiple days at distinct locations, providing multiple viewpoints of the same environment.
• Diverse Sensor Integration: It incorporates RGB-D data, precise GPS with geospatial heading, and dense LiDAR annotations, facilitating advanced technological applications in C-SLAM processes.
• Controlled Trajectory Overlap: Trajectory designs offer varied overlap levels allowing for the critical assessment of data association strategies amidst different environmental conditions and observational redundancies.
• Explicit LiDAR Annotations: Dense semantic labeling within LiDAR scans serves as a ground truth benchmark, supporting the thorough evaluation of segmentation and recognition algorithms.

Implications and Future Prospects

The implications and potential of the CU-Multi dataset stretch across both theoretical and practical domains. On a theoretical level, it provides a structured framework for advancing understanding within multi-robot perception alignment strategies and collaborative map generation methodologies. Practically, it affords the opportunity to improve reliability and resilience in robotic systems tasked with complex and collaborative missions.

Looking forward to developments in AI, the CU-Multi dataset sets a precedent for comprehensive evaluation methods, encouraging the gradual, yet decisive movement from single trajectory validation approaches towards more interconnected and dynamic robotic operation assessments. Expanding towards the integration of heterogeneous robotic platforms and diverse sensor configurations can further enrich the dataset’s applicability and practical deployment.

The release of CU-Multi signifies a forward step in the pursuit of robust multi-robot cooperation, paving the way for innovative solutions in collaborative data processing and interoperability scaling, fostering advancements in shared robotics intelligence.