XYZPIT/260826_pick_up_the_bread_bag

Pick Up the Bread Bag

Dataset Description

This dataset contains teleoperated bread-bag pickup demonstrations collected using the Galaxea R1 Lite robot.

In each episode, the robot picks up a bread bag from a conveyor-belt setup. The dataset is stored in the LeRobot dataset format and contains synchronized multi-view RGB videos, robot observations, and action commands.

The dataset contains a single task instruction:

pick up the bread bag

Dataset Summary

Item	Value
Robot	Galaxea R1 Lite
Collection method	Teleoperation
Number of episodes	100
Number of frames	57,272
Number of videos	400
Number of task instructions	1
Frames per second	31 FPS

Task Distribution

Task instruction	Episodes
pick up the bread bag	100
Total	100

Camera Observations

Each episode contains four synchronized RGB camera streams.

Feature	Camera view	Resolution
observation.images.head_rgb	Left head camera	1280 × 720
observation.images.head_right_rgb	Right head camera	1280 × 720
observation.images.left_wrist_rgb	Left wrist camera	640 × 360
observation.images.right_wrist_rgb	Right wrist camera	640 × 360

The videos are encoded using AV1 at 31 FPS.

Robot Observations

The dataset includes robot proprioceptive observations such as:

• Left and right arm joint positions
• Left and right arm joint velocities
• Left and right gripper states
• Left and right end-effector poses
• Chassis state
• Torso state
• Chassis and torso velocities
• IMU measurements

Actions

The action features include:

• Left and right arm joint targets
• Left and right gripper targets
• Chassis velocity commands
• Torso velocity commands

Intended Use

This dataset is intended for research on robot imitation learning and vision-based manipulation, particularly for learning bread-bag pickup behavior in a conveyor-belt environment.

Limitations

• The demonstrations were collected using a single Galaxea R1 Lite robot.
• The data were collected in a single workspace and conveyor-belt configuration.
• The dataset contains only one task instruction.
• The dataset has limited variation in objects and environmental conditions.
• Performance learned from this dataset may not directly generalize to different objects, conveyor configurations, lighting conditions, backgrounds, or robot embodiments.