Tiny Robotics Dataset and Benchmark for Continual Object Detection

Francesco Pasti,Riccardo De Monte,Davide Dalle Pezze,Gian Antonio Susto,Nicola Bellotto

2024-09-25

Abstract:Detecting objects in mobile robotics is crucial for numerous applications, from autonomous navigation to inspection. However, robots are often required to perform tasks in different domains with respect to the training one and need to adapt to these changes. Tiny mobile robots, subject to size, power, and computational constraints, encounter even more difficulties in running and adapting these algorithms. Such adaptability, though, is crucial for real-world deployment, where robots must operate effectively in dynamic and unpredictable settings. In this work, we introduce a novel benchmark to evaluate the continual learning capabilities of object detection systems in tiny robotic platforms. Our contributions include: (i) Tiny Robotics Object Detection (TiROD), a comprehensive dataset collected using a small mobile robot, designed to test the adaptability of object detectors across various domains and classes; (ii) an evaluation of state-of-the-art real-time object detectors combined with different continual learning strategies on this dataset, providing detailed insights into their performance and limitations; and (iii) we publish the data and the code to replicate the results to foster continuous advancements in this field. Our benchmark results indicate key challenges that must be addressed to advance the development of robust and efficient object detection systems for tiny robotics.

Robotics,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper aims to address the issue of continuous object detection by micro-robots in different environments. Specifically, micro-robots (such as the Tiny Robotics platform) are often subject to strict limitations in size, power, and computational capacity, making it difficult for them to run and adapt to complex object detection algorithms. However, in real-world deployments, these robots must be able to operate effectively in dynamic and unpredictable environments, which requires strong adaptive capabilities.

To tackle this challenge, the authors propose a new dataset and benchmarking framework—TiROD (Tiny Robotics Object Detection). This dataset was collected by a small mobile robot in 5 different environments, including both indoor and outdoor scenes, and covers 13 different object categories. The dataset is used to test the adaptability of object detectors across different domains and categories. Additionally, the authors evaluate the performance of several state-of-the-art real-time object detectors combined with different continual learning strategies and provide detailed performance analysis.

The main contributions of the paper include:
1. Proposing the TiROD dataset for testing the adaptability of different object detectors in various environments.
2. Evaluating the performance of two advanced real-time object detectors (NanoDet and Yolo v8) combined with different continual learning strategies on this dataset.
3. Releasing the source code for all experimental methods, including three continual learning object detection implementations that were not previously available online.

Through this work, the authors hope to advance the development of efficient and robust incremental object detection systems in the field of micro-robots, enabling their effective deployment in dynamic real-world environments.