README.md

---
dataset_info:
  features:
  - name: frame_index
    dtype: int64
  - name: episode_index
    dtype: int64
  - name: index
    dtype: int64
  - name: timestamp
    dtype: float32
  - name: next.done
    dtype: bool
  - name: next.success
    dtype: bool
  - name: action
    sequence: float32
    length: 6
  - name: observation.images.wrist.rgb
    dtype: video_frame
  - name: observation.images.base.rgb
    dtype: video_frame
  - name: observation.state
    sequence: float32
    length: 6
  - name: tcp_pose_rotvec
    sequence: float32
    length: 6
  splits:
  - name: train
    num_bytes: 4994773
    num_examples: 22373
  download_size: 2464205
  dataset_size: 4994773
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: mit
task_categories:
- robotics
---

A [`LeRobot`](https://github.com/huggingface/lerobot) -formatted dataset of 101 demonstrations for a 'planar push' task with a UR5e robot. 
The robot has a cylindrical end-effector and has to push an object to the target location. Once done, the robot needs to move to the corner of the workspace.

#TODO: add video 


The target location is fixed and marked with an orange square. Both the initial joint configuration of the robot and the pose of th
The dataset was collected using a [Gello](https://wuphilipp.github.io/gello_site/) arm to teleoperate the robot.


The robot's motion is contrained to remain top-down and on a height of z=1. The agent controls the 6 joints of the robot arm by providing absolute joint coordinates.

The observations are
 - 640x480 RGB image from the wrist camera (realsense)
 - 640x480 RGB image from a third-person camera (realsense)
 - the joint configuration

The robot is operated at 10Hz and each episode is succesful. There is no max duration.


Collecting these 100 episodes took about 1 hour.