Loco-Manipulation With Quadruped Robots: Modeling, Task Taxonomy, and Control Methods - A Critical Review

Loco-manipulation-the tight, simultaneous integration of locomotion and manipulation-has rapidly expanded the autonomy and usefulness of quadruped robots equipped with arms. This survey introduces a unified modeling-tasks-controls triad that explicitly links 1) floating-base and contact modeling assumptions; 2) loco-manipulation task regimes; and 3) controller families, and uses this structure to critically evaluate the literature through evidence-based trade-off maps, failure-mode synthesis, and reported quantitative anchors enabling consistent cross-paradigm comparison beyond existing reviews. Recent advances are synthesized across model-based inverse-dynamics Whole-Body Control (WBC), optimization-based whole-body Model Predictive Control (MPC), learning-based Reinforcement/Imitation Learning (RL/IL), and emerging hybrid architectures augmented by impedance and force-control layers. Grounded in an updated survey landscape and a refined task taxonomy, the triad is used to derive 'when-to-use-what' decision matrices spanning interaction type, terrain, payload, coordination, and sensing. High-fidelity simulation and digital-twin pipelines are highlighted as practical enablers for benchmarking, reproducibility, and reliable sim-to-real transfer. Using the proposed six-axis task taxonomy as the organizing lens, the hardest-to-transfer regimes consistently couple contact-rich end-effector interaction, uneven or uncertain footholds, and perception latency or occlusion. By consolidating fundamental trade-offs across control families and outlining research directions toward robust hybrid designs that couple predictive constraint handling with data-driven adaptation, this review charts a clear trajectory for deploying quadruped loco-manipulation in complex real-world environments.