{"paper":{"title":"AssemLM: A Spatial Reasoning Multimodal Large Language Model for Robotic Assembly","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"AssemLM integrates point clouds into a multimodal LLM via a specialized encoder to predict accurate 6D poses for robotic assembly.","cross_cats":[],"primary_cat":"cs.RO","authors_text":"Chenjia Bai, Huazhe Xu, Jicong Ao, Jinbin Qiao, Ouyang Lu, Shuang Qiu, Yu-Gang Jiang, Zhi Jing","submitted_at":"2026-04-10T05:43:39Z","abstract_excerpt":"Spatial reasoning is a fundamental capability for embodied intelligence, especially for fine-grained manipulation tasks such as robotic assembly. Recent methods based on vision-language models (VLMs) largely rely on coarse 2D perception and struggle to perform accurate reasoning over complex 3D geometry. To address this limitation, we propose AssemLM, a spatial multimodal large language model for robotic assembly that integrates assembly manuals, point clouds, and textual instructions to predict task-critical 6D assembly poses with explicit geometric understanding. To bridge raw 3D perception "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"AssemLM achieves state-of-the-art performance in 6D pose reasoning across diverse assembly scenarios. Furthermore, real-robot evaluations show that our model can support fine-grained and multi-step assembly execution in real-world settings.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the specialized point cloud encoder successfully captures fine-grained geometric and rotational features which integrate effectively with the multimodal language model to produce accurate 3D spatial reasoning that generalizes to real robots.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"AssemLM uses a specialized point cloud encoder inside a multimodal LLM to reach state-of-the-art 6D pose prediction for assembly tasks, backed by a new 900K-sample benchmark called AssemBench.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"AssemLM integrates point clouds into a multimodal LLM via a specialized encoder to predict accurate 6D poses for robotic assembly.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e97f4fb389520af028c1340d6d51effda8ad86072260a396ac92c21591946e05"},"source":{"id":"2604.08983","kind":"arxiv","version":2},"verdict":{"id":"11a06097-b50f-4c8c-8939-54249a68bbdb","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T18:19:37.665077Z","strongest_claim":"AssemLM achieves state-of-the-art performance in 6D pose reasoning across diverse assembly scenarios. Furthermore, real-robot evaluations show that our model can support fine-grained and multi-step assembly execution in real-world settings.","one_line_summary":"AssemLM uses a specialized point cloud encoder inside a multimodal LLM to reach state-of-the-art 6D pose prediction for assembly tasks, backed by a new 900K-sample benchmark called AssemBench.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the specialized point cloud encoder successfully captures fine-grained geometric and rotational features which integrate effectively with the multimodal language model to produce accurate 3D spatial reasoning that generalizes to real robots.","pith_extraction_headline":"AssemLM integrates point clouds into a multimodal LLM via a specialized encoder to predict accurate 6D poses for robotic assembly."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.08983/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"026899e177bb977d6df1403293a491ada9876e0988f47a3e32063cdd68caca2a"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}