Q-CoMove: Differentiable Quantum Circuit Priors for Coordinated Motion in Multi-Component Embodied Systems
Abstract
Multi-component embodied robots must coordinate the base, arm, camera, gripper, and local environment under shared spatial and temporal constraints. We present Q-CoMove, a coordinated motion framework that models this platform as a coupled dynamical system. Q-CoMove uses a differentiable circuit-style prior implemented with the TorchQuantum simulator on classical hardware: component states are encoded into a five-qubit register, cross-component dependence is parameterized by a 79-gate simulated circuit, and 25 measured statistics are projected into a coordination latent that conditions a physics-residual dynamics model and a joint action generator. We position this module as a compact structured coordination bottleneck rather than a claim of quantum advantage or physical necessity. Across parameter-matched Transformer and GNN baselines, residual-dynamics and joint MPC baselines, ablations, and robustness stress tests in simulation, Q-CoMove improves coordination metrics with bounded inference cost. All empirical results are limited to simulated environments, and the current study does not isolate the circuit prior against a structure-matched classical bottleneck.