Learning Generalizable Robot Skills from Demonstrations in Cluttered\n Environments

2018/08/01 by Muhammad Asif Rana, Mustafa Mukadam, Rana, Muhammad Asif +7
Engineering · Computer Science · #Robot Manipulation and Learning #Reinforcement Learning in Robotics #Adversarial Robustness in Machine Learning

paper · pdf · doi:10.48550/arxiv.1808.00349

Abstract

Learning from Demonstration (LfD) is a popular approach to endowing robots\nwith skills without having to program them by hand. Typically, LfD relies on\nhuman demonstrations in clutter-free environments. This prevents the\ndemonstrations from being affected by irrelevant objects, whose influence can\nobfuscate the true intention of the human or the constraints of the desired\nskill. However, it is unrealistic to assume that the robot's environment can\nalways be restructured to remove clutter when capturing human demonstrations.\nTo contend with this problem, we develop an importance weighted batch and\nincremental skill learning approach, building on a recent inference-based\ntechnique for skill representation and reproduction. Our approach reduces\nunwanted environmental influences on the learned skill, while still capturing\nthe salient human behavior. We provide both batch and incremental versions of\nour approach and validate our algorithms on a 7-DOF JACO2 manipulator with\nreaching and placing skills.\n

Related

• Learning from Few Demonstrations with Frame-Weighted Motion Generation
• How Can Everyday Users Efficiently Teach Robots by Demonstrations?
• Interactive Human-in-the-loop Coordination of Manipulation Skills Learned from Demonstration
• Learning Skills from Demonstrations: A Trend from Motion Primitives to Experience Abstraction
• Enabling Robots to Identify Missing Steps in Robot Tasks for Guided Learning from Demonstration