A U-turn involves a short loop, two to five segments in length, makes a sharp bend between the two secondary structures that it connects. The bend is somewhere around 180 degrees, making it a "u-turn".
U-turns are closely related to beta turns and beta hairpins. Beta turns are specific examples of U-turns, where the loop portion is only two segments long. A beta hairpin is a common arrangement of two sheets usually involves a U-turn. (The Greek letter beta in these names indicate they involve "beta-sheets", which are just "sheets" in Foldit.)
The Koga & Koga paper discussed similar types of turns, consisting of two to five loop segments. The loop patterns identified in Koga & Koga became the basis of Ideal Loops in Foldit. See design structures and Koga & Koga technical supplement for more.
U-turns often score poorly, but they may allow other, larger areas to score very well. For example, a U-turn may connect two well-positioned sheets. The sheets may share many hydrogen bonds, and have many hydrophobic segments hidden deep within the protein. Attempting to improve the score of the U-turn is likely to reduce the score of the sheets.
Beta-turns are discussed in A neural network method for prediction of beta-turn types in proteins using evolutionary information.
There are two web servers that use the methods from the article to predict beta turns: