Meta-Reinforcement Learning for Robotic Industrial Insertion Tasks
Authors:
- 24 October 2020
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 9728-9735
Abstract
Robotic insertion tasks are characterized by contact and friction mechanics, making them challenging for conventional feedback control methods due to unmodeled physical effects. Reinforcement learning (RL) is a promising approach for learning control policies in such settings. However, RL can be unsafe during exploration and might require a large amount of real-world training data, which is expensive to collect. In this paper, we study how to use meta-reinforcement learning to solve the bulk of the problem in simulation by solving a family of simulated industrial insertion tasks and then adapt policies quickly in the real world. We demonstrate our approach by training an agent to successfully perform challenging real-world insertion tasks using less than 20 trials of real-world experience.Keywords
All Related Versions
Funding Information
- Siemens
- Office of Naval Research
This publication has 23 references indexed in Scilit:
- Multi-Robot Assembly Strategies and MetricsACM Computing Surveys, 2018
- Domain randomization for transferring deep neural networks from simulation to the real world2017
- Deep reinforcement learning for high precision assembly tasks2017
- Learning from the hindsight plan — Episodic MPC improvement2017
- Mastering the game of Go with deep neural networks and tree searchNature, 2016
- Localization and manipulation of small parts using GelSight tactile sensing2014
- MuJoCo: A physics engine for model-based control2012
- Search strategies for peg-in-hole assemblies with position uncertainty2002
- Quasi-Static Assembly of Compliantly Supported Rigid PartsJournal of Dynamic Systems, Measurement, and Control, 1982
- Force Feedback Control of Manipulator Fine MotionsJournal of Dynamic Systems, Measurement, and Control, 1977