Aiming　at　the　problem　that　the　joints　were　easily　destroyed　by　the　impact　torque　during　the　process　of　dual-arm　space　robot　on-orbit　capturing　satellite，a　compliance　control　algorithm　based　on　reinforcement　learning　was　proposed．First，the　dynamic　models　of　dual-arm　space　robot　and　the　satellite　system　before　capture　were　derived　by　the　Lagrange　approach　and　Newton-Euler　method．Then，based　on　the　impulse　theorem，the　geometrical　conditions　and　Newton＇s　third　law，the　dynamic　model　of　the　hybrid　system　of　space　robot　and　the　satellite　after　capture　was　established．Considering　the　unstable　hybrid　system，a　buffer　compliance　reinforcement　learning　controller　was　proposed．　The　real-time　optimization　control　strategy　was　generated　by　interaction　with　dynamic　environment，which　could　realize　optimal　control　of　the　stabilization　phase．Numerical　simulation　results　show　that　the　proposed　control　strategy　can　reduce　the　impact　torque　by　54.2%　at　most，while　ensuring　the　joint　impact　torque　to　be　limited　to　a　safe　range　during　stabilization　phase，which　can　avoid　damage　from　the　impact　torque　for　the　joint.　©　2024　Huazhong　University　of　Science　and　Technology.　All　rights　reserved.