{"paper":{"title":"Hydra-MDP: End-to-end Multimodal Planning with Multi-target Hydra-Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Hydra-MDP trains an end-to-end planner by distilling knowledge from both human demonstrations and rule-based experts into a multi-head decoder that outputs diverse trajectories.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Jan Kautz, Jose M. Alvarez, Kailin Li, Shihao Wang, Shiyi Lan, Yishen Ji, Yu-Gang Jiang, Zhenxin Li, Zhiding Yu, Zhiqi Li, Ziyue Zhu, Zuxuan Wu","submitted_at":"2024-06-11T06:18:26Z","abstract_excerpt":"We propose Hydra-MDP, a novel paradigm employing multiple teachers in a teacher-student model. This approach uses knowledge distillation from both human and rule-based teachers to train the student model, which features a multi-head decoder to learn diverse trajectory candidates tailored to various evaluation metrics. With the knowledge of rule-based teachers, Hydra-MDP learns how the environment influences the planning in an end-to-end manner instead of resorting to non-differentiable post-processing. This method achieves the $1^{st}$ place in the Navsim challenge, demonstrating significant i"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"This method achieves the 1st place in the Navsim challenge, demonstrating significant improvements in generalization across diverse driving environments and conditions.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The multi-head decoder can simultaneously absorb conflicting signals from human and rule-based teachers without mode collapse or degraded performance on any single metric.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Hydra-MDP uses multi-teacher distillation and a multi-head decoder to learn diverse, metric-specific trajectories in an end-to-end autonomous-driving planner, winning the Navsim challenge.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Hydra-MDP trains an end-to-end planner by distilling knowledge from both human demonstrations and rule-based experts into a multi-head decoder that outputs diverse trajectories.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"de0c17d258b240ad85c5a3c16695a2db5e4cfb1acc8ecbb1ce5a4525f905cc2b"},"source":{"id":"2406.06978","kind":"arxiv","version":4},"verdict":{"id":"8a072b7e-d671-4586-894f-e8acd65f7dfd","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T23:08:25.962349Z","strongest_claim":"This method achieves the 1st place in the Navsim challenge, demonstrating significant improvements in generalization across diverse driving environments and conditions.","one_line_summary":"Hydra-MDP uses multi-teacher distillation and a multi-head decoder to learn diverse, metric-specific trajectories in an end-to-end autonomous-driving planner, winning the Navsim challenge.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The multi-head decoder can simultaneously absorb conflicting signals from human and rule-based teachers without mode collapse or degraded performance on any single metric.","pith_extraction_headline":"Hydra-MDP trains an end-to-end planner by distilling knowledge from both human demonstrations and rule-based experts into a multi-head decoder that outputs diverse trajectories."},"references":{"count":19,"sample":[{"doi":"","year":2024,"title":"Quad: Query-based in- terpretable neural motion planning for autonomous driving","work_id":"1a045cde-fbd3-4c5d-a59a-342f16668d6a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"NuPlan: A closed-loop ML-based planning benchmark for autonomous vehicles","work_id":"b16aace6-deff-4546-b333-bcb7c9c07cdb","ref_index":3,"cited_arxiv_id":"2106.11810","is_internal_anchor":true},{"doi":"","year":null,"title":"VADv2: End-to-End Vectorized Autonomous Driving via Probabilistic Planning","work_id":"e7670f83-e1e1-41e7-86eb-39477a3a10b2","ref_index":4,"cited_arxiv_id":"2402.13243","is_internal_anchor":true},{"doi":"","year":2022,"title":"Transfuser: Imita- tion with transformer-based sensor fusion for autonomous driving","work_id":"8b9d9ecc-fbc9-4be4-88eb-e6170c75b588","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Navsim: Data-driven non-reactive autonomous vehicle simulation","work_id":"ab2ad404-1d69-4f8c-bf71-ceae9114f626","ref_index":6,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":19,"snapshot_sha256":"8c42e9a988fece09949b512fd17a08917c7b60c5907f38794a4ae7d315b9a618","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"c62df76cd2204039c5969cebd20513310a5be001de94ef67328d611a0164f5b7"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}