Statistical methods that adapt to unknown population structures are attractive due to both numerical and theoretical advantages over their non-adaptive counterparts. In this work we contribute to adaptive modelling of functional regression, where challenges arise from the infinite dimensionality of functional predictor in the underlying space. We are interested in the scenario that the predictor process lies in a potentially nonlinear manifold that is intrinsically infinite-dimensional and embedded in an infinite-dimensional functional space. By a novel approach built upon local linear manifold smoothing, we achieve a polynomial rate of convergence that adapts to the level of noise/sampling contamination and intrinsic manifold dimension with an interesting phase transition phenomenon when functional trajectories are observed intermittently with noise, in contrast to the logarithmic rate in nonparametric functional regression literature. We demonstrate that the proposal enjoys favorable finite sample performance relative to commonly used methods via simulated and real data examples.