HTMD is build around adaptive sampling, i.e. use on-the-fly information from the current data to decide where to restart new simulations. This in practice works really well with speed-up of over an order of magnitude compared to standard sampling methods. Adaptive sampling replaces to a large extent any need for biased sampling in a much safer way than biasing. Also it does not require to have very good reaction coordinates between it is building its own reaction space on-the-fly while sampling. It also integrate very well with Markov state models.
Relevant papers to read: