A　new　approach　of　fast　collision-free　motion　planning　for　underactuated　robots　based　on　genetic　algorithm　is　proposed.　The　collision　avoidance　problem　is　formulated　and　solved　as　a　position-based　force　control　problem.　Virtual　generalized　force　representing　the　intrusion　of　the　arm　into　the　obstacle　dangerous　zone　is　computed　in　real　time　using　a　virtual　spring-damper　model.　The　partly　stable　controllers　are　adopted　and　the　energy　based　fitness　function　is　built,　then　the　best　switching　sequence　of　partly　stable　controllers　is　obtained　by　genetic　algorithm.　Because　the　proposed　method　does　not　make　any　hypothesis　about　the　degree　of　freedom,　it　can　be　used　without　modification　for　arms　with　a　large　number　of　degree　of　freedom.　At　last,　numerical　simulations　which　are　carried　on　the　planar　3R　underactuated　robots　show　the　effectiveness　of　the　proposed　method.