<p>Human-robot collaboration (HRC) requires robot motions that are task-efficient, predictable, physically safe, and ergonomically compatible with human partners in shared workspaces. Existing methods often treat demonstration reproduction and ergonomic optimization separately, which may lead to motions that either follow demonstrated task geometry without sufficient comfort consideration or improve ergonomics at the expense of task fidelity. This paper proposes a human-centric Composite Field (CF) motion planning framework that unifies demonstration-informed task representation and ergonomic awareness in a continuous task-space formulation. A human-demonstrated distance field (HDDF) is constructed from cooperative demonstrations to encode task geometry, while a task space ergonomic field (TSEF) is obtained by projecting a configuration space ergonomic field (CSEF) through human forward kinematics to represent spatial ergonomic comfort. By fusing these two fields into a single composite potential, the proposed framework provides an interpretable and tunable trade-off between trajectory adherence and ergonomic improvement. A gradient-flow-based motion generator follows the negative gradient of the CF to produce smooth human-centric references, which are executed by a model predictive impedance controller for compliant and coordinated robot motion. Simulations and real-world experiments in physical human-robot interaction and collaborative box transportation show that the proposed method preserves demonstrated motion characteristics while reducing ergonomic strain and muscle activation. Compared with Point-to-Point, HDDF-only, and TSEF-only baselines, the CF planner achieves a more balanced performance between task fidelity and human comfort.</p>