In　this　paper,　a　simple　end-to-end　obstacle　avoidance　method　is　introded　for　manipulators.　First　of　all,　the　2D　images　in　the　workspace　are　used　to　uniformly　describe　the　characteristics　of　obstacles　in　the　3D　space.　After　that,　a　model-free　reinforcement　learning　algorithm　DrQ-v2　is　used　to　train　the　obstacle　avoidance　strategy,　which　directly　outputs　the　joint　angles　to　avoid　the　obstacles　autonomously　and　accurately　in　the　joint　space.　Finally,　the　simulation　results　demonstrate　that　the　proposed　end-to-end　simple　control　method　is　more　effective　and　convenient　in　handling　obstacle　avoidance　tasks　in　complex　dynamic　environments.　©　2023　IEEE.