Acquisition of a Multi-Armed Robot System for Automated Experimentation in DoD-Relevant Robot Learning Research
Abstract
Research into robot learning requires increasingly larger amounts of data. Modern learning algorithms require large amounts of data to train, and the need to evaluate and compare different methods further multiples these data requirements. To keep up with these demands, robot learning experiments will need to become more automated. We therefore propose a multi-armed robotic system for autonomously performing experiments for DoD-relevant research into robot manipulation learning. The arms will work together to perform different steps of tasks, including resetting objects for each other. The proposed system will allow the robots to learn more complex tasks and collect more data autonomously. The system will reduce the need for human intervention for running the experiments, e.g., to reset the robot~s environment between attempts. The system will also be used to teach students how to createautonomous learning experiments. The proposed system will thus have a high impact on both research and research-related teaching.The proposed system will support DoD-relevant research and establish new research capabilities. The robot platform will allow us to research deep learning methods for representing objects in the world based on physical interactions with them. Similar to a human, the robots will be able to alter their surroundings to learn how to perform skills in different situations. With the proposedsystem, we will develop methods for robots to autonomously choose the most informative settings for learning. We will also develop methods for transferring the skills learned on the proposed system to mobile robots, such that the skills can be employed more quickly in the field. The system~s software will be built on a learning architecture that will be developed as part of a research project that was recently recommended for funding by the Office of Naval Research.In this manner, future research projects will automatically become integrated into this life-long learning framework.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 23, 2019
- Source ID
- N000141912208
Entities
People
- Oliver Kroemer
Organizations
- Massachusetts Institute of Technology
- Office of Naval Research
- United States Navy