On　the　basis　of　trajectory　imitation　learning,　aiming　at　the　problem　of　the　poor　learning　effect　caused　by　the　fluctuations　of　the　single　demonstration　data,　this　paper　presents　a　kind　of　imitation　learning　method　based　on　multiple　demonstrations.　The　multi　constraint　optimization　idea　is　introduced　for　trajectory　optimization.　Multiple　demonstrations　data　are　coded　by　GMM　separately,　and　the　probability　interval　parameters　are　set　as　the　optimization　constraints.　The　intersection　of　the　constraint　conditions　is　to　obtain　the　interval　overlap　probability.　The　probability　of　superior　data　is　higher,　and　the　probability　of　inferior　data　is　lower.　Parameter＇s　set　for　multi　constrained　optimal　trajectory　are　solved　based　on　Gaussian　distribution　product　form,　and　then　obtain　a　multi　constrained　optimal　trajectory　through　GMR.　This　method　reduces　the　probability　of　bad　data　in　the　demonstration.　It　effectively　avoids　the　fluctuation　of　the　reproduction　trajectory　caused　by　the　inferior　demonstration　data.　The　simulation　results　show　the　effectiveness　of　the　method.　©　2017　IEEE.