A key challenge towards the goal of multi-part assembly tasks is finding\nrobust sensorimotor control methods in the presence of uncertainty. In contrast\nto previous works that rely on a priori knowledge on whether two parts match,\nwe aim to learn this through physical interaction. We propose a hierarchical\napproach that enables a robot to autonomously assemble parts while being\nuncertain about part types and positions. In particular, our probabilistic\napproach learns a set of differentiable filters that leverage the tactile\nsensorimotor trace from failed assembly attempts to update its belief about\npart position and type. This enables a robot to overcome assembly failure. We\ndemonstrate the effectiveness of our approach on a set of object fitting tasks.\nThe experimental results indicate that our proposed approach achieves higher\nprecision in object position and type estimation, and accomplishes object\nfitting tasks faster than baselines.\n