📈 GH Trending Explorer 🎲 Surprise me

Search: corl · Language: Python · clear

23Star / xianyu-super-butler

闲鱼超级管家是一款全面升级的xy自动发货平台,自动滑块、发货、评价、要花、擦亮通通稳定支持,时刻维护!时刻更新!坚决保证使用世界上最强 AI模型(6 Astra\6 Sol 、ops 5)vibecoding! 群在最后面,不想部署用在线的xy.corleom.com(免费、最新、安全,托管账号AI自动过滑块)

★ 1.3k Python today
NVlabs / GRAIL

🤖 GRAIL is a fully digital data-generation pipeline that synthesizes humanoid loco-manipulation data from world models. 🌟 CoRL 2026 (Oral)

★ 576 Python 5d ago
noitom-robotics / AdaPT

[CoRL 2026] The official implementation of "AdaPT: Towards Professional Tennis Styles for Humanoid Robots with Adaptive Motion Planning and Tracking."

★ 107 Python 2mo ago
Bariona / FACT

[CoRL 2026] Code for "FACT: Failure-Aware Causal Training for World‑Action Models"

★ 105 Python 1mo ago
mx-liu6 / OptiGeo

[CORL2026] Official implementation of the paper: "OptiGeo: Efficient Monocular Geometry for Embodied Perception in Optically Challenging Scenes"

★ 94 Python 1mo ago
amazon-far / PRISM

[CoRL 2026] Counterfactual Video Generation Enables Scalable Humanoid Loco-Manipulation

★ 82 Python 11d ago
wyhuai / HumanX

[CoRL 2026] HumanX: Toward Agile and Generalizable Humanoid Interaction Skills from Human Videos

★ 77 Python 11d ago
3587jjh / HuRo

HuRo: Robotizing Human Videos for Scalable VLA Pretraining (CoRL 2026)

★ 59 Python 29d ago
ambench / ambench

AM-Bench: a modular simulation suite and benchmark for aerial manipulation policy learning (CoRL 2026)

★ 55 Python 1mo ago
sharpa-robotics / WM-Craftnet

[CoRL'2026 Spotlight] WM-Craftnet: World Synesthesia Model for Generalizable and Robust Dexterous In-Hand Manipulation

★ 32 Python 10d ago
AureleoPKU / DeCAL

Official Implementation for “DeCAL: Towards Physically-Grounded Dexterous Vision-Language-Action Models via Contact-Aware Latent Co-Imagination” (CoRL 2026)

★ 27 Python 1mo ago
Ingrid789 / OmniContact

[🎉 CORL 2026] OmniContact: Chaining Meta-Skills via Contact Flow for Generalizable Humanoid Loco-Manipulation

★ 25 Python 12d ago
ASGuard-UCI / safedrive-vla

[CoRL 2026] SafeDriveVLA: Navigation-Conditioned World Model Dreaming for Conflict-Aware End-to-End Autonomous Driving