pollen-robotics/reachy-doing-things

Reachy Doing Things Dataset

Overview

The Reachy Doing Things Images Dataset consists of images captured from the perspective of the Reachy humanoid robot. These images were taken during teleoperation sessions, providing a unique view of the environment as perceived by the robot during manipulation tasks. The images were captured with a RGBD camera mounted on the shoulder of the robot.

Purpose

This dataset is primarily aimed at testing and validating the performance of vision tools integrated into the pollen-vision library. Currently, it serves as a validation suite for object detection, object segmentation, and image tagging algorithms.

While it is not intended for training models at this stage, it provides valuable real-world data to refine and improve these vision tools. The images in the dataset are not annotated.

Content

• Non annotated ego-centric images captured during teleoperation sessions with Reachy

Future Updates

While the dataset is currently limited in size, ongoing efforts are underway to expand its content. Future updates will include additional images and annotations, making it a more comprehensive resource for vision research and development.

Contact

For any inquiries or feedback regarding the dataset, please contact contact@pollen-robotics.com.

---

license: apache-2.0 dataset_info: features:

• name: image dtype: image
• name: label dtype: class_label: names: '0': croissant '1': kitchen-counter '2': serving-cup '3': whiteboard splits:
• name: train num_bytes: 3760395.0 num_examples: 23 download_size: 3762376 dataset_size: 3760395.0 configs:
• config_name: default data_files:
  • split: train path: data/train-*

---