This repository contains the code for the paper "E2CL: Exploration-based Error Correction Learning for Embodied Agents"
-
🎯 Novel Exploration Framework: Introduces E2CL, an innovative approach that leverages exploration-induced errors and environmental feedback to enhance embodied agents' performance
-
🤖 Self-Correction Capability: Enables agents to learn from mistakes through teacher-guided and teacher-free explorations, developing robust self-correction abilities
-
🚀 Superior Performance: Demonstrates significant improvements over traditional methods in the VirtualHome environment, achieving better environment alignment and task execution