The review, published open access on 13 September 2026, focuses specifically on dual-arm robotic multi-peg-in-hole assembly, abbreviated DA-MPiH. This is the variant of the problem where two robot arms must cooperate to manipulate a part — often a large, flexible, or awkwardly shaped component — and align it with multiple mating features at once. The authors frame the task as fundamentally contact-rich: it involves multi-point contact coupling, bimanual closed-chain constraints, error propagation, and sensing uncertainty. In plain terms, when two arms grip a single workpiece, they form a kinematically closed loop in which the forces each arm applies are not independent. If one arm drifts by a fraction of a millimeter, the other must absorb the resulting internal stress, or the entire assembly will bind. This is precisely the regime where classical position control fails and where intelligence — in perception, reasoning, and control — must take over.

To build their evidence base, the team conducted a genuinely systematic search. They queried four major databases — the Web of Science Core Collection, Scopus, IEEE Xplore and arXiv — for literature published between 2008 and July 2026, supplementing the search with backward citation tracking. After deduplication and screening following the PRISMA protocol, the standard methodology for systematic reviews in medicine and now increasingly in engineering, 191 studies made the final cut. Each study was coded by robot configuration, peg-hole scale, validation setting, and relevance to the dual-arm multi-peg task. That coding scheme matters, because it allowed the authors to ask a question that most narrative reviews in robotics never answer rigorously: how many of these papers actually test their methods on the full dual-arm, multi-peg problem, rather than on a simplified proxy?

The answer is the review’s most sobering finding. While learning-based perception and control demonstrably improve a robot’s adaptation under uncertain contact conditions, the overwhelming majority of the 191 studies address single-arm or single-peg tasks. Direct experimental evidence that integrates dual-arm coordination with multi-peg constraints remains limited. This is not merely an academic quibble. Techniques that work brilliantly for a single rigid peg — reinforcement learning policies trained in simulation, force-guided search strategies, learned contact-state estimators — do not automatically transfer when a second arm enters the picture and the part acquires multiple simultaneous contact interfaces. The closed-chain constraint between the two arms introduces internal forces that have no counterpart in single-arm assembly, and the review argues that these internal forces are systematically under-addressed in the current literature.

The review organizes the AI-enabled toolbox into several interlocking layers. The first is system composition: what sensors, actuators and computational architectures dual-arm assembly cells actually deploy. The second is cooperative and contact-state modeling, the mathematical machinery for reasoning about which surfaces of the peg are touching which surfaces of the hole at any instant. Contact-state reasoning is the intellectual heart of the problem, because a robot that knows its contact state can predict whether pushing harder will advance the assembly or jam it irreversibly. The third layer covers target recognition and search — the pre-contact strategies by which the robot localizes holes with cameras and plans exploratory trajectories, often combining deep-learning vision models with spiral or force-guided search patterns to compensate for residual localization error.

The fourth layer, compliant control, is where the review draws its sharpest technical distinctions. Passive compliance relies on mechanical elasticity, such as remote center of compliance devices, that physically absorb alignment errors without any computation. Active compliance uses force and torque feedback to modulate the robot’s motion in real time, letting it respond to contact forces within milliseconds. Learning-based compliance, the newest and fastest-growing category, uses reinforcement learning, imitation learning and related techniques to acquire insertion strategies that would be prohibitively difficult to hand-engineer. The authors find that learning-based approaches genuinely improve adaptation under uncertainty — a policy trained with domain randomization can tolerate part tolerances and fixture variations that would defeat a fixed controller — but they also caution that these gains come with costs that the field rarely reports honestly.

That reporting problem is the review’s second major critique. Performance metrics and training costs are documented so inconsistently across studies that strict cross-study comparison is effectively impossible. One paper may report success rates on a specific peg-hole clearance ratio with a specific sensor suite; another may report only qualitative demonstrations. Training a reinforcement learning policy can require millions of simulated episodes or thousands of physical trials, yet few papers quantify the computational budget, the sim-to-real gap, or the failure modes encountered during transfer. Without standardized reporting, a laboratory manager hoping to deploy dual-arm assembly on a production line has no rigorous way to judge which published method would survive contact with their own parts, tolerances and cycle-time requirements. The review explicitly calls for standardized DA-MPiH benchmarks to fix this.

The authors also identify challenges in sensor fusion, interpretability and safe learning that cut across the entire field. Multimodal sensing — combining vision, force-torque data, tactile arrays and joint encoders — promises the richest contact-state estimates, but fusing these streams reliably under the noise and latency of real hardware remains unsolved. Interpretability matters because an assembly policy that fails unpredictably on a factory floor is worse than a weaker but transparent controller. And safe skill transfer — moving a policy learned in simulation, or on one robot, onto another without dangerous force spikes — is a prerequisite for any industrial adoption, since a two-meter robot arm applying uncontrolled forces to a machined aluminum housing can destroy thousands of dollars of parts in a fraction of a second.

Looking forward, the review lays out a research agenda with four priorities: multimodal contact estimation, internal-force-aware compliant control, safe skill transfer, and standardized benchmarks. The internal-force priority deserves particular emphasis, because it is the feature that most cleanly separates dual-arm assembly from everything that came before it. A controller that treats the two arms as independent single-arm agents will generate fighting forces through the workpiece; a controller that explicitly models and regulates the internal stress within the closed chain can exploit bimanual manipulation for what it is actually good at — handling large, heavy or compliant parts that no single arm could manage. Whether reinforcement learning architectures can internalize this constraint, or whether it must be built in through constrained optimization and hybrid force-position control, is one of the field’s most interesting open questions.

The significance of this work extends well beyond robotics conferences. Multi-peg-in-hole assembly stands in for an entire class of contact-rich manipulation tasks — connector mating in electronics, fastener insertion in aerospace, joinery in construction — that still resist automation and still consume enormous amounts of skilled human labor. The Guizhou University team’s systematic accounting of 191 studies makes clear that the AI community has built powerful components: vision systems that localize holes, policies that wiggle pegs home, controllers that yield gracefully to unexpected contact. What it has not yet built, in most cases, is the integrated, dual-arm, multi-peg system that industry actually needs, validated on real hardware with reproducible metrics. The review’s message to the field is essentially a challenge: stop publishing single-peg proxies, start reporting training costs, and build the benchmarks that will let the next generation of bimanual assembly robots be compared, improved and, ultimately, deployed.

Subject of Research: Artificial intelligence methods for dual-arm robotic multi-peg-in-hole assembly

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

Article References: Zhang, W., Yuan, Q., Qu, P., Jia, W., & Zhang, Y. (2026). Artificial intelligence for dual-arm robotic multi-peg-in-hole assembly: a review. Artificial Intelligence Review. https://doi.org/10.1007/s10462-026-11705-4

Image Credits: AI Generated

DOI: 10.1007/s10462-026-11705-4

Keywords: dual-arm robotics, multi-peg-in-hole assembly, artificial intelligence, reinforcement learning, compliant control, contact-state modeling, bimanual manipulation, robotic assembly, systematic review, sensor fusion, sim-to-real transfer, industrial automation

The scale of the challenge becomes clear when the current state of the art is examined honestly. Existing lunar and Mars missions still depend overwhelmingly on pre-programmed command sequences and teleoperation from Earth, an approach that imposes severe latency penalties and leaves robots nearly helpless when conditions deviate from expectations. The study identifies five core technical problems that must be solved before large-scale lunar development becomes practical: multi-task high-precision manipulation, autonomous navigation and obstacle avoidance, self-learning interactive collaboration, adaptation to extreme environments, and high-reliability long-duration operation. Each of these problems is magnified by the lunar setting itself, where abrasive dust, extreme temperature swings, radiation, and complex illumination near the poles combine to degrade sensors, mechanisms, and electronics. The authors contend that systematically integrating artificial intelligence, deep learning, large-scale models, and embodied intelligence into lunar robot design has become the critical bottleneck for the entire enterprise of lunar exploration and development.

To organize this vast technical landscape, the research team proposes a developmental roadmap grounded in the phased milestones of human lunar activity. In the first stage, the programmed robot era, machines operate through pre-programmed commands and ground teleoperation, a mode suited to the unmanned scientific exploration phase that has defined lunar robotics to date. The second stage, the intelligent robot era, is subdivided into three hierarchical levels: weak intelligence, intelligence, and general intelligence. These levels correspond respectively to the unmanned lunar research station phase, the lunar base phase, and ultimately a lunar community phase in which fully autonomous operations become the norm. The capability demands evolve in step with these phases, from basic mobility and single-arm manipulation in early surveys, to multi-task autonomous operations at a research station, to complex assembly, construction, and human-robot collaboration at a base, and finally to fully intelligent autonomous behavior across an entire lunar settlement.

At the heart of the paper is a technological framework the authors describe through an anatomical metaphor: a brain, a cerebellum, and a body. The brain serves as the robot’s core command center, receiving external information, planning actions, and generating instructions. It encompasses multimodal perception and information fusion, autonomous mission planning and decision-making, large-model reasoning, path planning and obstacle avoidance, health monitoring, cloud computing and intelligent chips, and human-robot and swarm collaboration. The cerebellum handles fine-grained regulation of motion, translating high-level commands into real-time adjustments based on the robot’s current state; its technologies include autonomous navigation and localization in complex terrain, multi-arm collaborative compliant control, and reinforcement learning-based motion control. The body is the executor, and it demands high-torque long-life modular joints, high-mobility multi-modal locomotion mechanisms, versatile end effectors, high-specific-energy distributed power systems, wireless power transmission, and environment-adaptive design. Together, these three layers define what the authors call an intelligence plus new energy empowerment philosophy for lunar robotics.

One of the most demanding technical problems the team tackled is visual perception in the lunar south pole region, where low sun angles produce long shadows, harsh contrast, and terrain surfaces with repetitive, low-texture features that confound conventional stereo matching. To address this, the researchers developed a visual perception algorithm built on a lightweight deep network. In preliminary experiments, the algorithm achieved stable feature extraction and matching under varying illumination conditions and accurately recovered depth information in scenes with repetitive textures, enabling high-precision mapping and localization even in the difficult lighting regimes expected near the poles. The authors emphasize that this capability lays a direct foundation for autonomous navigation on the lunar surface, since a robot that cannot reliably build a map of its surroundings and locate itself within that map cannot plan a safe path, avoid hazards, or execute any of the construction and maintenance tasks that future missions will require.

Manipulation is the second pillar of the team’s prototype verification work. The researchers established a ground verification platform comprising an equivalent manipulator, an end quick-change mechanism, a controller, and a multifunctional tool kit, and employed a nonlinear compliant control method to achieve high-precision position tracking and contact force buffering at the end effector. Through 50 repeated measurements of position accuracy and 30 repeated measurements of orientation accuracy, the locking precision of the end quick-change device was verified in all three directions to meet the requirements for fine manipulation. This matters because lunar construction will demand that robots swap tools, grasp irregular objects, assemble structures, and interact safely with both equipment and humans, all while absorbing contact forces that would otherwise damage rigid mechanisms or destabilize the robot itself. Compliant control, in other words, is the difference between a machine that can only push and pull and one that can genuinely build.

Mobility across unstructured lunar terrain forms the third pillar. The team’s wheel-leg hybrid locomotion subsystem uses four independent deployable mechanisms, each chain offering three degrees of freedom, allowing the robot to switch between wheeled travel and legged climbing as conditions demand. In a vehicle body lifting experiment, the robot stably raised its chassis from a squatting posture through coordinated wheel-leg motion in approximately 10 seconds, with motor current and torque feedback remaining within safe operating ranges and no jamming at the joint pivots. The mechanism effectively climbed a slope of 20.2 degrees and traversed a 28.3 millimeter step, a height exceeding twice the wheel diameter, demonstrating favorable adaptability to the rocks, craters, and loose regolith that characterize the lunar surface. These results suggest that hybrid locomotion could give future lunar robots the versatility to handle terrain that would defeat purely wheeled rovers.

Based on this body of analysis and experimentation, the study closes with three development recommendations. First, the authors call for a unified technical framework and a consensus on technology classification for lunar intelligent robotics, clearly defining the functional scope and technical metrics for each intelligence level so that progress can be measured and compared. Second, they advocate a phased advancement strategy: near-term breakthroughs in weak-intelligence technologies to build practical engineering application capabilities, followed by medium-to-long-term development of full intelligence and general intelligence. Third, they urge the formulation of international industry standards covering overall system design, interface specifications, and environmental adaptability, so that robots built by different nations can interoperate and collaborate on shared lunar infrastructure. Without such standards, they warn, the fragmented development of national systems could undermine the cooperative vision that programs like the International Lunar Research Station are meant to embody.

The significance of this work extends well beyond any single robot design. By mapping mission phases to intelligence levels, and intelligence levels to concrete technologies in perception, decision-making, motion control, and actuation, the study offers a systematic reference framework for the planning, technological development, and engineering application of robotic systems in the construction of lunar research stations and lunar bases. It also frames a broader transition now underway across the space sector, in which the tools of modern artificial intelligence, from deep learning to large-scale models to embodied intelligence, are being pulled out of terrestrial laboratories and pressed into service in deep space. If the roadmap the authors describe proves accurate, the robots that build humanity’s first permanent footholds on the Moon will not be remotely piloted machines awaiting instructions from Earth, but autonomous partners capable of working, adapting, and surviving alongside the explorers they serve. The groundwork for that transition, the study suggests, is being laid today.

Subject of Research: Development and key technologies of intelligent robotic systems for lunar exploration and base construction

Article Title: Prospect and research progress of lunar intelligent robot technology

Article References: Prospect and research progress of lunar intelligent robot technology. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: lunar robots, artificial intelligence, autonomous navigation, lunar base, embodied intelligence, compliant control, wheel-leg locomotion, visual perception, human-robot collaboration, International Lunar Research Station, Artemis program, space robotics