/bih_dataset

Benchmarking In-Hand Manipulation Dataset

This repo contains dataset for benchmarking in-hand manipulation on different robot platforms. More details are available at https://robot-learning.cs.utah.edu/project/benchmarking_in_hand_manipulation

Repo structure:

  • The hand poses dataset exists in the folder dataset, a readme exists that discuses the dataset and how to use it.
  • Initial and desired contact mesh regions are available at the provided URL.
  • Additionally, results on robot platforms are available inside the folder named results.
  • The subfolders in results have the name ROBOTNAME_lvl_NO, where ROBOTNME refers to robot platform and NO refers to the level of benchmarking in-hand manipulation.

Benchmarked Methods

Level I

  1. Relaxed-Rigidity
  2. Relaxed-position
  3. Relaxed-position-orientation
  4. Point Contact with Friction
  5. IK-Rigid

Level III

  1. Dexterous Manipulation Graphs

To Contribute:

The provided dataset works sufficiently well for human sized robotic hands. We encourage the research community to run their in-hand manipulation scheme with our dataset on the YCB objects set and submit their results as a pull request.