Work

Learning 1000 Robot Manipulation Tasks in a Day (Code and Data)

Year: 2025

Type: article

Abstract: In this repository, we are providing the code to replicate all results published in the Science Robotics paper: "Learning 1000 Tasks in a Day". Together with the code, you will find tutorials on how to run and train the different robot learning methods. Additionally, we also share the weights of the trained models we deemed useful and generalisable to various robot setups. Finally, we include data collected directly on the robot. These are robot rollouts that were labelled as either success or failure, and for which we share data such as RGB-D images, joint angles, gripper states, and all other pertinent information that can be valuable to the community for model training.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.17334510

Authors: Pietro, Vitiello, Dreczkowski, Kamil

Institution: Imperial College London

Language: English

Topic: Robot Manipulation and Learning

Subfield: Control and Systems Engineering

Field: Engineering

Domain: Physical Sciences

Open Access status: green

Funders: Royal Academy of Engineering, Engineering and Physical Sciences Research Council

Awards: G432953325, DTP 2224 Imperial College London, Efficient Robot Learning of Contact-Rich Tasks from Non-Expert Demonstrations