Artificial intelligence for dual-arm robotic multi-peg-in-hole assembly: a review

Dual-arm robotic multi-peg-in-hole (DA-MPiH) assembly is a contact-rich manipulation task involving multi-point contact coupling, bimanual closed-chain constraints, error propagation, and sensing uncertainty. Artificial intelligence offers new opportunities for improving perception, contact-state reasoning, search, and adaptive compliant control; however, the extent to which existing methods have been validated directly on DA-MPiH tasks remains unclear. This review systematically analyzes the physical foundations and AI-enabled methods for DA-MPiH assembly. Literature published from 2008 to July 2026 was retrieved from the Web of Science Core Collection, Scopus, IEEE Xplore, and arXiv, supplemented by backward citation searching. Following deduplication and PRISMA-based screening, 191 studies were included and coded by robot configuration, peg-hole scale, validation setting, and relevance to DA-MPiH assembly. The review synthesizes system composition, cooperative and contact-state modeling, target recognition and search, and passive, active, and learning-based compliance. The evidence shows that learning-based perception and control improve adaptation under uncertain contact, but most studies address single-arm or single-peg tasks; direct experimental evidence integrating dual-arm coordination and multi-peg constraints remains limited. Inconsistent performance and training-cost reporting further prevents strict cross-study comparison. Future research should prioritize multimodal contact estimation, internal-force-aware compliant control, safe skill transfer, and standardized DA-MPiH benchmarks.