Nanobodies exhibit antigen binding affinities of the same order as those of antibodies, which, along with their small size and unique structural characteristics, makes them well-suited for therapeutic and diagnostic applications. The lack of coevolutionary signals in nanobody-antigen complexes together with the broad complementary determining region 3 loop (CDR3) conformational space poses a challenge for predicting the 3D structure of those complexes with computational modelling and artificial intelligence-based methods. In this context, physics-based information-driven docking can provide an alternative solution. This study evaluates the state-of-the-art of machine learning-based methods for nanobody structure prediction and benchmarks various HADDOCK workflows to model their interaction with antigens using different input nanobody ensembles and information scenarios. We propose an ensemble docking pipeline that achieves high success rates starting from nanobody structural models predicted by AlphaFold2 and ImmuneBuilder. Our results highlight the effectiveness of physics-based complex prediction of immune proteins when accurate input structures and sufficient information to guide the modelling are available.