randomConfiguration method errors out when loading a model that includes floating joints

[Zionshang]

Bug description

When loading a model that includes floating joints, the randomConfiguration method errors out.
The method:

  template<typename Scalar, int Options, template<typename, int> class JointCollectionTpl>
  typename PINOCCHIO_EIGEN_PLAIN_TYPE_NO_PARENS(
    (typename ModelTpl<Scalar, Options, JointCollectionTpl>::ConfigVectorType))
    randomConfiguration(const ModelTpl<Scalar, Options, JointCollectionTpl> & model)
  {
    return randomConfiguration<LieGroupMap, Scalar, Options, JointCollectionTpl>(model);
  }

The error

terminate called after throwing an instance of 'std::range_error'
  what():  non bounded limit. Cannot uniformly sample joint at rank 0

Expected behavior

The randomConfiguration method should correctly generate a random configuration without errors.

Reproduction steps

Steps to reproduce the behavior:

1. Load a model that includes floating joints.
2. Call the randomConfiguration method. The error occurred.
3. Set the floating joint to a fixed joint, then the error will disappear

Code

Model model;
pinocchio::urdf::buildModel(urdf_filename, model);
Eigen::VectorXd q = pinocchio::randomConfiguration(model);

[jcarpent]

By default, translation components of the in model.lowerPositionLimit and model.upperPositionLimit are set to infinity for floating joints. You need to manually set them up after loading your robot model to some relevant values for your application.