[Paper Review] Compare Contact Model-based Control and Contact Model-free Learning: A Survey of Robotic Peg-in-hole Assembly Strategies

Jing Xu, Zhimin Hou+2 more|arXiv (Cornell University)|Apr 10, 2019

Robot Manipulation and Learning93 references84 citations

TL;DR

The paper surveys robotic peg-in-hole assembly strategies, comparing traditional contact model-based control with two branches of contact model-free learning (learning from demonstrations and learning from environments), and discusses integration, challenges, and future directions.

ABSTRACT

In this paper, we present an overview of robotic peg-in-hole assembly and analyze two main strategies: contact model-based and contact model-free strategies. More specifically, we first introduce the contact model control approaches, including contact state recognition and compliant control two steps. Additionally, we focus on a comprehensive analysis of the whole robotic assembly system. Second, without the contact state recognition process, we decompose the contact model-free learning algorithms into two main subfields: learning from demonstrations and learning from environments (mainly based on reinforcement learning). For each subfield, we survey the landmark studies and ongoing research to compare the different categories. We hope to strengthen the relation between these two research communities by revealing the underlying links. Ultimately, the remaining challenges and open questions in the field of robotic peg-in-hole assembly community is discussed. The promising directions and potential future work are also considered.

Motivation & Objective

• Group existing peg-in-hole strategies into contact model-based and contact model-free categories.
• Analyze contact state recognition and compliant control within model-based approaches.
• Survey learning-from-demonstrations and reinforcement-learning-based methods in model-free approaches.
• Discuss links between traditional models and data-driven learning and identify open challenges.

Proposed method

• Review and classify literature on peg-in-hole assembly into two main families: contact model-based and contact model-free methods.
• Within contact model-based, analyze contact state recognition (analytical vs. statistical) and compliant control (low-level and high-level planning).
• Within contact model-free, separate learning from demonstrations (LFD) and learning from environments (LFE, including RL and model-based RL discussions).
• Compare methods using reported metrics such as success rate and computational time for contact state recognition techniques.
• Summarize interfaces between model-based and model-free paradigms to suggest hybrid approaches and future research directions.

Experimental results

Research questions

• RQ1 What are the main characteristics and limitations of contact model-based control versus contact model-free learning in peg-in-hole assembly?
• RQ2 How do learning-from-demonstrations and reinforcement-learning-based strategies compare in terms of robustness, data efficiency, and generalization?
• RQ3 What are the linking principles between traditional contact models and implicit model learning from demonstrations?
• RQ4 What open challenges and future directions emerge for integrating these strategies in real-world assembly?

Key findings

• Analytical contact state recognition can be sensitive to uncertainties and may not generalize well to new environments.
• Statistical methods (GMM, SVM) offer better generalization for contact state recognition but have trade-offs in accuracy and computation.
• HMMs can incorporate temporal information to improve state transition recognition in peg-in-hole assembly.
• Learning-from-demonstrations and reinforcement-learning-based methods provide flexible handling of variability and unstructured environments, with LFD including DMPs, GMMs, and HMMs as encoding strategies.
• Table II reports comparative performance for recognition methods, e.g., 94.4% success rate for GMM/DSM-GMM, 64.2% for SVM-based methods, and 60.7% for SGB (with various trade-offs).
• Model-based RL discussions indicate potential benefits when fusing prior knowledge with learning for peg-in-hole tasks.