Embodied Self Supervised Learning

Embodied self-supervised learning (EMSSL) aims to train robots to learn complex tasks through interaction with their environment without relying on extensive human-labeled data. Current research focuses on developing EMSSL frameworks for inverse kinematics modeling in robotics, leveraging techniques like conditional generative models and active learning to improve efficiency and precision. These methods often incorporate 3D scene understanding and symbolic reasoning to enable more robust and generalizable learning. The ultimate goal is to enable robots to autonomously acquire skills and adapt to new environments, significantly advancing robotics and artificial intelligence.

Papers

October 14, 2024

Embodied Active Learning of Generative Sensor-Object Models
Allison Pinosky, Todd D. Murphey
Robot Person Generative Question PAC Bayesian Novel Object Sensing Model Embodied Self Supervised Learning

June 22, 2023

CEMSSL: A Unified Framework for Multi-Solution Inverse Kinematic Model Learning of Robot Arms with High-Precision Manipulation
Qu Weiming, Liu Tianlin, Luo Dingsheng
Unified Framework Inverse Kinematics Robot Arm Precise Manipulation Embodied Self Supervised Learning

March 8, 2023

Embodied Active Learning of Relational State Abstractions for Bilevel Planning
Amber Li, Tom Silver
Bilevel Optimization State Abstraction Neural Predicate Symbolic Planner Relational Algebra Embodied Self Supervised Learning

February 26, 2023

Embodied Self-Supervised Learning (EMSSL) with Sampling and Training Coordination for Robot Arm Inverse Kinematics Model Learning
Qu Weiming, Liu Tianlin, Wu Xihong, Luo Dingsheng
Open Sampling Inverse Kinematics Robot Arm Joint Training Embodied Self Supervised Learning

December 2, 2021

SEAL: Self-supervised Embodied Active Learning using Exploration and 3D Consistency
Devendra Singh Chaplot, Murtaza Dalal, Saurabh Gupta, Jitendra Malik, Ruslan Salakhutdinov
Active Learning Environment Exploration Perception Model 3D Consistency Robot Vision 3D Semantic Map Embodied Self Supervised Learning

Embodied Self Supervised Learning

Papers

Embodied Active Learning of Generative Sensor-Object Models

CEMSSL: A Unified Framework for Multi-Solution Inverse Kinematic Model Learning of Robot Arms with High-Precision Manipulation

Embodied Active Learning of Relational State Abstractions for Bilevel Planning

Embodied Self-Supervised Learning (EMSSL) with Sampling and Training Coordination for Robot Arm Inverse Kinematics Model Learning

SEAL: Self-supervised Embodied Active Learning using Exploration and 3D Consistency