{"id":"arxiv:2610.40356","kind":"preprint","title":"JEDI: Joint-Embedding Demonstration Compression for In-Context Imitation Learning","authors":["Jiazhao Liang","Zhewen He","Shuaihang Yuan","Hao Huang","Yi Fang"],"author_count":5,"year":2026,"published":"2026-10-29","venue":null,"cited_by":null,"doi":null,"arxiv_id":"2610.40356","nct_id":null,"source_url":"https://arxiv.org/abs/2610.40356","archive_url":"https://space-kitty.com/papers/arxiv:2610.40356","updated":"2026-10-29","pmid":null,"pmcid":null,"openalex_id":null,"pdf_url":"https://arxiv.org/pdf/2610.40356","abstract":"In-context imitation learning (ICIL) adapts robot policies through test-time demonstrations, but long multimodal trajectories constrain how many examples fit in context. We ask what a compressed demonstration should preserve. We propose JEDI, which compresses demonstrations into latent slots trained to predict the query episode's next-stage keyframe in a frozen video-JEPA space. Task and process counterfactuals help the slots preserve execution details, while the predictor selects demonstrations from a bank. Post-training with keypoint tokens enables human-hand conditioning on robots. Experiments on LIBERO and real robots show that JEDI supports many-shot context; LIBERO objective controls show improved success and demonstration following over action-only compression. These findings support query-future prediction as a training objective for preserving execution-relevant information in compressed demonstrations.","abstract_license":"CC0-1.0 (arXiv descriptive metadata)","license":null,"is_open_access":true,"retracted":false,"peer_reviewed":false,"categories":["cs.RO"],"details":null,"provenance":"arxiv","attribution":"Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.","topics":[{"slug":"robotics-embodied","name":"Robot Learning & Embodied AI","domain_slug":"ai"}],"cited_in_dossiers":[]}