Heterogeneous Demonstration

Heterogeneous demonstration learning focuses on enabling robots and AI agents to learn effectively from diverse and imperfect human demonstrations, a crucial step towards more user-friendly and adaptable systems. Current research emphasizes developing algorithms that can handle variations in demonstration quality and style, often employing inverse reinforcement learning or preference learning within frameworks that build and adapt policy mixtures from multiple demonstrations. This research is significant because it addresses a major bottleneck in deploying reinforcement learning in real-world applications, where obtaining perfectly consistent expert demonstrations is often impractical or impossible. Improved algorithms in this area will facilitate more efficient and robust robot training and human-robot interaction.

Papers

February 2, 2024

Inverse Reinforcement Learning by Estimating Expertise of Demonstrators
Mark Beliaev, Ramtin Pedarsani
Imitation Learning Inverse Reinforcement Learning Noisy Demonstration Suboptimal Demonstration Heterogeneous Demonstration

October 22, 2023

Learning to Discern: Imitating Heterogeneous Human Demonstrations with Preference and Representation Learning
Sachit Kuhar, Shuo Cheng, Shivang Chopra, Matthew Bronars, Danfei Xu
LeArning Abstract Representation Learning Policy Learning Preference Feedback Human Demonstration Offline Imitation Heterogeneous Demonstration

August 18, 2023

Enhancing Agent Communication and Learning through Action and Language
Hugo Caselles-Dupré, Olivier Sigaud, Mohamed Chetouani
LeArning Abstract Human Language Agent Smith Action Feature Agent Communication Human Communication Language Instruction Heterogeneous Demonstration

March 20, 2023

A Survey of Demonstration Learning
André Correia, Luís A. Alexandre
Reinforcement Learning Timely Survey Robotics Domain Demonstration Data Heterogeneous Demonstration

September 24, 2022

Fast Lifelong Adaptive Inverse Reinforcement Learning from Demonstrations
Letian Chen, Sravan Jayanthi, Rohan Paleja, Daniel Martin, Viacheslav Zakharov, Matthew Gombolay
Inverse Reinforcement Learning Noisy Demonstration Fast Adaptation Adaptive Robot Heterogeneous Demonstration

February 14, 2022

Strategy Discovery and Mixture in Lifelong Learning from Heterogeneous Demonstration
Sravan Jayanthi, Letian Chen, Matthew Gombolay
Mixture Component Lifelong Learning Task Specific Reward Adaptive Robot Strategy Discovery Heterogeneous Demonstration