[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-80848":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":16,"stars30d":17,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":18,"rankGlobal":9,"rankLanguage":9,"license":19,"archived":20,"fork":20,"defaultBranch":21,"hasWiki":22,"hasPages":20,"topics":23,"createdAt":9,"pushedAt":9,"updatedAt":24,"readmeContent":25,"aiSummary":26,"trendingCount":15,"starSnapshotCount":15,"syncStatus":16,"lastSyncTime":27,"discoverSource":28},80848,"CLOVER","WilliamXuanYu\u002FCLOVER","WilliamXuanYu","CLOVER, a Closed-LOop Value Estimation and Ranking framework for end-to-end driving planning.",null,"Python",42,3,36,1,0,2,6,1.81,"Apache License 2.0",false,"main",true,[],"2026-06-12 02:04:07","# CLOVER\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"fig\u002Fimage1.png\" alt=\"Pipeline\">\n\u003C\u002Fp>\n\nEnd-to-end autonomous driving planners are commonly trained by imitating a single logged trajectory, yet they are evaluated by rule-based planning metrics that measure safety, feasibility, progress, and comfort. This creates a training-evaluation mismatch: trajectories close to the logged path may still violate planning rules, while alternative trajectories farther from the demonstration can remain valid and high-scoring. The mismatch is especially limiting for proposal-selection planners, whose performance depends on both candidate-set coverage and scorer ranking quality. We propose **CLOVER**, a **C**losed-**LO**op **V**alue **E**stimation and **R**anking framework for end-to-end driving planning. CLOVER first expands single-trajectory imitation into set-level proposal coverage by constructing evaluator-filtered pseudo-expert trajectories. It then performs conservative closed-loop self-distillation: a trajectory-level scorer is fitted to true evaluator sub-scores on generated proposals, while the generator is refined toward teacher-selected top-k and vector-Pareto proposal targets with stability regularization. We also analyze when an imperfect scorer can improve the generator, showing that scorer-mediated refinement is reliable under local scorer accuracy, conservative updates, and selected-set enrichment.\n\nPaper: `https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.15120`\n\n## TODO\n\n- [x] Release paper\n- [x] Release inference code, scripts, and ckpt\n- [x] Release preview training scripts\u003Csup>[1]\u003C\u002Fsup>\n- [ ] Release official training code\n- [ ] Release pseudo-expert trajectory generation code and NAVSIM-v2 evaluation scripts\n\n> **Note [1]**\n> To facilitate early community discussion and reproduction, we release this preview version of the training scripts first. This preview may still contain unfinished details, deprecated interfaces, or fixed-path assumptions. These issues will be cleaned up in the formal release. The epoch schedule may also differ slightly from the final paper setup. In the current stage-2 preview we default to 30 alternating cycles (30 x 2 epochs in total). Empirically, the best checkpoint is often reached around 20 to 30 epochs, but iterative alternating training can occasionally be unstable, and an early score drop during the first several epochs is normal. We therefore keep a longer default schedule in the preview release.\n\n## Diversity Visualization\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"fig\u002Fdiversity_visualization_appendix.png\" alt=\"Diversity visualization appendix\">\n\u003C\u002Fp>\n\n## Releases\n\n- Checkpoints and release assets: `https:\u002F\u002Fgithub.com\u002FWilliamXuanYu\u002FCLOVER\u002Freleases`\n- DINOv2 ViT-S backbone weights: `https:\u002F\u002Fhuggingface.co\u002Ftimm\u002Fvit_small_patch14_reg4_dinov2.lvd142m\u002Ftree\u002Fmain`\n- Stage-1 pseudo-expert trajectory package: `https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1oNTv5Pe-naw_i81rqaKk8KIs0VcUqGZ-?usp=drive_link`\n\n## Installation\n\n```bash\nconda create -n clover python=3.8\nconda activate clover\npip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu121\npip install -r requirements.txt\npip install -e \u002Fpath\u002Fto\u002Fnuplan-devkit\npip install -e .\n```\n\nIf you prefer to use the vendored `nuplan-devkit` copy in this repository instead of an external checkout:\n\n```bash\npip install -e .\u002Fnuplan-devkit\n```\n\n## Documentation\n\n- Training guide: [docs\u002Ftraining.md](docs\u002Ftraining.md)\n- Inference guide: [docs\u002Finference.md](docs\u002Finference.md)\n\n## Public Entrypoints\n\n- Train metric cache: `python navsim\u002Fplanning\u002Fscript\u002Frun_train_metric_caching.py`\n- Stage-1 training: `bash scripts\u002Frun_training_multi_expert.sh`\n- Stage-2 training: `bash scripts\u002Frun_training_stage2_vector_pareto_alternating.sh`\n- NAVSIM-v1 evaluation: `bash scripts\u002Feval_multi_expert_navtest.sh`\n","CLOVER 是一个用于端到端驾驶规划的闭环价值评估和排序框架。其核心功能在于通过构建评估器筛选的伪专家轨迹，将单一轨迹模仿扩展为集合级别的提案覆盖，并通过保守的闭环自蒸馏来优化生成器与评分器，从而提高规划的安全性、可行性和舒适度。技术上，CLOVER 采用 Python 编写，支持 DINOv2 ViT-S 作为骨干网络，适用于自动驾驶系统中的路径规划场景，特别是在需要平衡候选集覆盖率和评分质量的情况下。该框架旨在解决训练-评估不匹配的问题，使规划器能够生成更符合实际需求的驾驶策略。","2026-06-11 04:02:33","CREATED_QUERY"]