This article describes a single-query probabilistic roadmap planner, called Small Tree, for hyper-redundant manipulators. The planner incrementally builds two solution paths from small alternating road-maps rooted at the two input configurations, until a connection linking both paths is found. In this manner, the solution path grows from both input queries. The process has the advantage of continuously replacing the two initial input queries, whose original connectivity in configuration space may have been difficult, with new target configurations. The method also has the ability to initiate the manipulator motion before the complete solution is found by using the partial solution path connected to the initial configuration.