Learning New Tasks via Reusable Skills: Skill-Compositional Experts for Embodied Continual Learning

· Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

The Skill-Compositional Experts (SCE) framework is proposed to address catastrophic forgetting and feature drift in Embodied Continual Learning (ECL) for robots. ECL aims for robots to continually acquire new manipulation tasks while retaining prior behaviors under closed-loop control, but struggles with structured skill reuse. SCE tackles this by first employing Compositional Skill Grounding (CSG) to decompose task demonstrations into reusable skills, forming a skill base. Subsequently, Dual Execution-and-Transition Experts (DETE) facilitate new task learning through skill composition, with distinct branches managing skill execution and supporting transitions between skills for coherent behavior. Experiments on LIBERO benchmarks and real-world manipulation tasks demonstrate that SCE consistently improves both retention of learned behaviors and overall task performance. The framework was published on 2026-06-14.

Key takeaway

For robotics engineers designing embodied continual learning systems, SCE provides a structured framework to overcome catastrophic forgetting. You should consider implementing skill decomposition via Compositional Skill Grounding to create reusable skill bases. This approach, combined with Dual Execution-and-Transition Experts for managing skill execution and transitions, can significantly improve retention and overall task performance in new manipulation task acquisition. Evaluate its effectiveness on your specific real-world manipulation tasks.

Key insights

SCE enables robots to learn new tasks continually by composing reusable skills, mitigating catastrophic forgetting in embodied learning.

Principles

Method

SCE uses Compositional Skill Grounding (CSG) to build a skill base from demonstrations, then Dual Execution-and-Transition Experts (DETE) for new task learning via skill composition.

In practice

Topics

Best for: Research Scientist, AI Scientist, Robotics Engineer

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.