This　paper　tackles　the　task　of　estimating　the　state　of　an　object　in　a　robotic　hand.　The　state　of　the　object　includes　its　shape　and　posture,　which　is　critically　important　for　robotic　in-hand　manipulation.　However,　in-hand　objects　have　self-occlusion,　making　it　challenging　to　perceive　their　complete　shape　and　posture.　To　address　this　challenge,　this　work　proposed　a　point-clouds　processing　framework　to　achieve　shape　completion　and　pose　estimation　of　the　in-hand　objects.　Firstly,　the　input　point　cloud　are　segmented　based　region　growing　algorithm　to　obtain　the　points　belonging　to　the　target　object.　Then,　we　design　a　neural　network　with　the　auto-encoder　structure　to　perform　shape　completion　and　6D　pose　estimation　of　the　in-hand　object.　The　latent　feature　of　the　network　is　used　to　regress　the　6D　pose,　i.e.,　position　and　orientation,　of　the　object.　The　effectiveness　of　the　proposed　framework　is　evaluated　by　comparison　experiment　and　real-word　experiment.　Experimental　results　show　that　our　approach　achieves　significantly　high　accuracy　in　the　shape　completion　and　pose　estimation　of　robotic　in-hand　objects.　©　2023,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.