http://localhost Daily picks in robot learning — RL / IL / VLA across cobot arms, bimanual, mobile manipulators, and humanoids. Wed, 27 May 2026 23:04:13 +0000 https://github.com/Jiaaqiliu/Awesome-VLA-Robotics github:Jiaaqiliu/Awesome-VLA-Robotics Mon, 23 Mar 2026 18:25:44 +0000 Jiaaqiliu/Awesome-VLA-Robotics is a curated survey-style resource for Vision-Language-Action models in robotics, organizing papers, models, datasets, benchmarks, and related tools across manipulation, navigation, mobile manipulation, HRI, planning, humanoids, and other embodied-AI settings. It is useful as a field map for tracking how VLA systems extend VLMs with robot action generation, including common building blocks such as vision encoders, LLM-based language understanding, action decoders or policies, and modality-alignment mechanisms. — Tags: #VLA #Manipulator #MobileManipulator #Humanoid #Survey — ★ 473 — Score 51.678 https://github.com/YanjieZe/awesome-humanoid-robot-learning github:YanjieZe/awesome-humanoid-robot-learning Sun, 24 May 2026 09:18:32 +0000 A Paper List for Humanoid Robot Learning. — Tags: #Humanoid #Survey — ★ 2343 — Score 49.52 https://github.com/jonyzhang2023/awesome-embodied-vla-va-vln github:jonyzhang2023/awesome-embodied-vla-va-vln Mon, 25 May 2026 12:06:47 +0000 jonyzhang2023/awesome-embodied-vla-va-vln is a curated reading list for embodied AI work around vision-language-action models, vision-language navigation, and related multimodal learning. It is useful as a research map rather than a runnable system, helping track papers, models, datasets, and benchmarks across robot learning and embodied navigation. — Tags: #VLA #Other #Survey — ★ 3151 — Score 47.791 https://github.com/starVLA/starVLA github:starVLA/starVLA Fri, 22 May 2026 02:39:51 +0000 StarVLA: A Lego-like Codebase for Vision-Language-Action Model Developing — Tags: #VLA #Other — ★ 2567 — Score 45.711 https://github.com/NVlabs/ProtoMotions github:NVlabs/ProtoMotions Tue, 19 May 2026 23:08:12 +0000 GPU-accelerated ProtoMotions3 trains physically simulated humanoids on motion corpora, claiming AMASS-scale skill learning in 12 hours on 4 A100s. Distinctive pieces are PyRoki one-command retargeting, IsaacGym/Newton/MuJoCo sim-to-sim testing, and ONNX sim-to-real deployment for Unitree G1. — Tags: #VLA #Humanoid — ★ 1670 — Score 45.501 https://github.com/huggingface/lerobot github:huggingface/lerobot Mon, 25 May 2026 21:59:35 +0000 🤗 LeRobot: Making AI for Robotics more accessible with end-to-end learning — Tags: #Other — ★ 24337 — Score 45.03 https://github.com/OpenHelix-Team/VLA-Adapter github:OpenHelix-Team/VLA-Adapter Thu, 19 Mar 2026 14:09:37 +0000 VLA-Adapter provides code, checkpoints, and training/evaluation scripts for adapting vision-language-action models on LIBERO and CALVIN, with configs spanning 10GB consumer GPUs to 80GB accelerators. It also adds real-world ALOHA/Cobot Magic deployment support. — Tags: #RL #VLA #Bimanual — ★ 2164 — Score 45.013 https://github.com/isaac-sim/IsaacLab github:isaac-sim/IsaacLab Mon, 25 May 2026 12:07:03 +0000 Unified framework for robot learning built on NVIDIA Isaac Sim — Tags: #Other — ★ 7253 — Score 44.186 https://github.com/Denghaoyuan123/Awesome-RL-VLA github:Denghaoyuan123/Awesome-RL-VLA Mon, 18 May 2026 08:32:13 +0000 Curated RL-VLA reading list for robotic manipulation, organizing papers by offline, online, offline+online, and test-time RL regimes. It adds comparison tables for action type, reward sparsity, MF/MB status, sim/real validation, base VLA, and policy class, but no new algorithm. — Tags: #RL #VLA #Manipulator #Survey — ★ 711 — Score 42.769 https://github.com/unitreerobotics/unitree_sim_isaaclab github:unitreerobotics/unitree_sim_isaaclab Mon, 30 Mar 2026 03:10:40 +0000 Unitree’s Isaac Lab simulator runs G1 and H1-2 humanoid manipulation tasks such as pick-place, block stacking, and whole-body object moving, with configurations for grippers, Dex3, and Inspire hands. It mirrors the real robots’ DDS communication topics, making it useful for testing control code, collecting or replaying teleoperation data with xr_teleoperate, and validating policies in simulation before moving to hardware. — Tags: #Manipulator #Humanoid — ★ 471 — Score 42.609 https://github.com/leggedrobotics/pace-sim2real github:leggedrobotics/pace-sim2real Fri, 22 May 2026 12:57:05 +0000 PACE is a sim-to-real pipeline for legged robots that estimates actuator and joint dynamics using only standard joint encoder data, avoiding extra sensing or specialized identification hardware. The repo is aimed at making legged-robot simulation models match real hardware more closely, so policies trained in simulation transfer more reliably to physical robots. — Tags: #Other — ★ 490 — Score 41.605 https://github.com/NVlabs/GR00T-VisualSim2Real github:NVlabs/GR00T-VisualSim2Real Mon, 20 Apr 2026 17:32:28 +0000 GR00T-VisualSim2Real VIRAL Visual Sim-to-Real at Scale for Humanoid Loco-Manipulation DoorMan Opening the Sim-to-Real Door for Humanoid Pixel-to-Action Policy Transfer Overview This repository contains the application code for VIRAL (Visual Sim-to-Real for Humanoid Loco-Manipulation) and DoorMan . — Tags: #VLA #Manipulator #Humanoid — ★ 250 — Score 41.466 https://github.com/google-deepmind/mujoco_playground github:google-deepmind/mujoco_playground Tue, 19 May 2026 21:57:55 +0000 GPU-accelerated MuJoCo MJX environments for robot RL and sim-to-real let researchers train on classic control, quadruped/biped locomotion, and dexterous or non-prehensile manipulation tasks. Distinctive support includes JAX MJX plus MuJoCo Warp backends and vision via MJWarp Batch Renderer. — Tags: #RL #Manipulator #Humanoid — ★ 1953 — Score 41.26 https://github.com/rohanpsingh/LearningHumanoidWalking github:rohanpsingh/LearningHumanoidWalking Sun, 03 May 2026 01:05:26 +0000 Train and evaluate deep-RL humanoid locomotion policies in MuJoCo, with H1/JVRC walking, footstep-tracking, terrain, and cartpole test environments. Its value is released code behind three humanoid-walking papers: current-feedback/back-EMF randomization, planned-footstep walking, and compliant/uneven-terrain robustness. — Tags: #RL #Humanoid — ★ 1142 — Score 40.778 https://github.com/jonyzhang2023/awesome-humanoid-learning github:jonyzhang2023/awesome-humanoid-learning Mon, 16 Mar 2026 11:43:56 +0000 Curated index of humanoid and bipedal robot learning resources covering locomotion, manipulation, whole-body control, physics-based animation, robot models, papers, news, and related lists. Its value is aggregation and model metadata, not a new algorithm or benchmark. — Tags: #Manipulator #Humanoid #Survey — ★ 921 — Score 40.697 https://github.com/mithi/robotics-coursework github:mithi/robotics-coursework Mon, 11 May 2026 08:28:29 +0000 🤖 Places where you can learn robotics (and stuff like that) online 🤖 — Tags: #Other — ★ 4654 — Score 40.575 https://github.com/leggedrobotics/rsl_rl github:leggedrobotics/rsl_rl Tue, 19 May 2026 08:46:53 +0000 A fast and simple implementation of learning algorithms for robotics. — Tags: #Other — ★ 2624 — Score 40.421 https://github.com/Genesis-Embodied-AI/genesis-world github:Genesis-Embodied-AI/genesis-world Mon, 25 May 2026 17:38:35 +0000 A generative world for general-purpose robotics & embodied AI learning. — Tags: #Other — ★ 28844 — Score 40.158 https://github.com/unitreerobotics/unitree_rl_lab github:unitreerobotics/unitree_rl_lab Mon, 18 May 2026 07:46:41 +0000 IsaacLab-based RL environments let researchers train policies for Unitree Go2, H1, and G1-29dof robots, with standalone tasks and robot assets via USD or URDF. The repo also documents MuJoCo sim2sim validation and direct sim2real deployment, indicating integration rather than a new RL method. — Tags: #RL #Other — ★ 1030 — Score 39.845 http://arxiv.org/abs/2605.25477v1 arxiv:2605.25477 Mon, 25 May 2026 06:31:03 +0000 EXPO-FT fine-tunes pretrained vision-language-action policies with reinforcement learning in a way that preserves their useful priors while making online robot learning stable and sample-efficient. On challenging manipulation tasks like routing string lights and plugging them in, striking a pool ball into a pocket, and inserting a flower into a wine bottle, it reaches 30/30 successes on every evaluated task using an average of 19.1 minutes of robot data, outperforming both RL-from-scratch and prior VLA fine-tuning baselines. — Tags: #RL #VLA #Manipulator — Score 39.644 https://github.com/sou350121/VLA-Handbook github:sou350121/VLA-Handbook Mon, 25 May 2026 05:20:34 +0000 VLA-Handbook is a Chinese, practice-oriented learning and interview guide for algorithm engineers moving into Vision-Language-Action robotics. It narrows in on robotics-specific issues rather than general CV/NLP preparation, making it useful as a focused entry point for people who need to understand the concepts, workflows, and interview expectations around VLA systems. — Tags: #VLA #Other — ★ 240 — Score 39.485 https://github.com/DravenALG/awesome-vla-wam github:DravenALG/awesome-vla-wam Mon, 18 May 2026 15:32:39 +0000 Curated index of Vision-Language-Action and World Action Model robotics research, organized by VLA variants, WAM sources, policies, datasets, benchmarks, engines, and hardware. Its value is taxonomy and literature tracking; it does not introduce a model, dataset, or benchmark. — Tags: #VLA #Other #Survey — ★ 445 — Score 39.398 https://github.com/hzxie/DynamicVLA github:hzxie/DynamicVLA Sun, 03 May 2026 09:14:27 +0000 DynamicVLA provides training, inference, and Isaac Lab evaluation code for a vision-language-action policy aimed at dynamic object manipulation on the DOM benchmark. It bundles DOM data, 3D assets, synthetic generation, LeRobot conversion, and a pretrained checkpoint; no new model mechanism is detailed here. — Tags: #VLA #Manipulator — ★ 253 — Score 39.177 https://github.com/ARISE-Initiative/robosuite github:ARISE-Initiative/robosuite Sat, 09 May 2026 19:23:19 +0000 robosuite: A Modular Simulation Framework and Benchmark for Robot Learning — Tags: #Manipulator — ★ 2425 — Score 39.085 https://github.com/Farama-Foundation/Gymnasium-Robotics github:Farama-Foundation/Gymnasium-Robotics Wed, 20 May 2026 09:04:16 +0000 A collection of robotics simulation environments for reinforcement learning — Tags: #RL #Other — ★ 909 — Score 38.867 https://github.com/mujocolab/mjlab github:mujocolab/mjlab Mon, 25 May 2026 18:14:07 +0000 Isaac Lab API, powered by MuJoCo-Warp, for RL and robotics research — Tags: #RL #Other — ★ 2382 — Score 38.791 http://arxiv.org/abs/2603.15046v1 arxiv:2603.15046 Mon, 16 Mar 2026 09:57:45 +0000 AnoleVLA is a lightweight vision-language-action model using deep state space models for language-guided mobile manipulation. It targets object manipulation from vision and natural-language instructions; no code, benchmark, or robot-test details are given here. — Tags: #VLA #Manipulator #MobileManipulator — Score 38.646 https://github.com/AgibotTech/ACoT-VLA github:AgibotTech/ACoT-VLA Thu, 21 May 2026 04:07:05 +0000 ACoT-VLA is the official CVPR 2026 implementation for Action Chain-of-Thought, a method for vision-language-action models that makes robot policies reason through intermediate action-oriented steps rather than mapping perception and language directly to actions. The repo likely provides the training and evaluation code needed to reproduce the paper’s experiments, making it useful for researchers studying interpretable or structured reasoning in embodied robot control. — Tags: #VLA #Other — ★ 175 — Score 38.102 http://arxiv.org/abs/2603.18532v2 arxiv:2603.18532 Sat, 28 Mar 2026 07:03:02 +0000 Method for scaling RL fine-tuning of robot vision-language-action models via generative 3D simulation. The excerpt frames it against real-world VLA fine-tuning, but gives no numbers, benchmarks, code release, or robot-test details. — Tags: #RL #VLA #Other — Score 38.041 https://github.com/2toinf/X-VLA github:2toinf/X-VLA Wed, 06 May 2026 18:33:10 +0000 X-VLA provides pretrained and fine-tuned cross-embodiment VLA checkpoints using embodiment-specific soft prompts to steer one 0.9B Transformer policy across robot platforms. Released models cover LeRobot/server-client inference and benchmarks including LIBERO 98.1%, Google Robot 83.5% VM/76.4% VA, WidowX 95.8%, CALVIN 4.43, and RoboTwin2 70%. — Tags: #VLA #Manipulator — ★ 656 — Score 37.957 http://arxiv.org/abs/2605.27284v1 arxiv:2605.27284 Tue, 26 May 2026 17:01:10 +0000 FineVLA targets the mismatch between robot policies that need to follow execution-style instructions and datasets that usually only say the goal, by adding action-aligned language about details like which arm to use, approach direction, pose, color, and contact region. It builds a 47,159-trajectory human-verified dataset from 972,247 trajectories across 10 robot datasets, plus a 500-video benchmark and a robotics-specialized VLM annotator for scaling fine-grained labels. Policies trained with fine-grained plus raw goal instructions performed best, reaching 86.8%/82.5% in RoboTwin and 62.7/100 on real dual-arm manipulation, with especially large gains on steerable factors that goal-only language leaves unspecified. — Tags: #VLA #Manipulator #Bimanual — Score 37.735 https://github.com/fan-ziqi/rl_sar github:fan-ziqi/rl_sar Wed, 27 May 2026 04:53:42 +0000 rl_sar collects simulation-to-real reinforcement learning workflows for deploying locomotion policies across quadruped, wheeled, and humanoid robots. It is aimed at validating robot RL algorithms in simulation and carrying them through to physical hardware, making it useful as a practical bridge between training experiments and real-world deployment. — Tags: #RL #Humanoid — ★ 1314 — Score 36.907 https://github.com/iLearn-Lab/NeurIPS25-CogVLA github:iLearn-Lab/NeurIPS25-CogVLA Wed, 27 May 2026 08:47:56 +0000 CogVLA is a NeurIPS 2025 vision-language-action model that aligns robot control with cognition by using instruction-driven routing and sparsification, so different parts of the model are selectively activated depending on the task command. The repo appears to package the method for researchers interested in efficient VLA policies that can condition robot behavior on language while avoiding dense, one-size-fits-all computation. — Tags: #VLA #Other — ★ 182 — Score 36.416 https://github.com/ustcwhy/BitVLA github:ustcwhy/BitVLA Mon, 02 Mar 2026 08:59:24 +0000 BitVLA provides 1.58-bit/1-bit vision-language-action checkpoints and evaluation code for robotics manipulation, including VQA and LIBERO workflows. The 3.0B model reports 1.4GB memory use and 96.0 avg LIBERO success after VL/VLA pre-training, near OpenVLA-OFT’s 97.1 at 15.4GB. — Tags: #VLA #Manipulator — ★ 152 — Score 36.393 https://github.com/Farama-Foundation/Metaworld github:Farama-Foundation/Metaworld Sun, 17 May 2026 03:27:18 +0000 Meta-World is an open source benchmark for developing and evaluating multi-task and meta reinforcement learning algorithms for continuous control robotic manipulation environments, with various benchmarks to evaluate different aspects of reinforcement learning algorithms. — Tags: #RL #Manipulator — ★ 1821 — Score 36.079 http://arxiv.org/abs/2603.22760v1 arxiv:2603.22760 Tue, 24 Mar 2026 03:44:25 +0000 SG-VLA is a spatially grounded VLA method for mobile manipulation in household scenes. It targets layout reasoning, fine geometry, and continuous high-dimensional actions; no numbers, code, or robot-test details are given here. — Tags: #VLA #Manipulator #MobileManipulator — Score 35.904 http://arxiv.org/abs/2603.13707v2 arxiv:2603.13707 Tue, 17 Mar 2026 23:57:56 +0000 REFINE-DP fine-tunes diffusion policies with reinforcement learning for humanoid loco-manipulation. It targets coordinated motion planning and stable whole-body execution for complex, long-horizon tasks; code, benchmark, and real-robot details aren’t specified. — Tags: #RL #IL #Manipulator #Humanoid — Score 35.698 https://github.com/OpenBMB/DeepThinkVLA github:OpenBMB/DeepThinkVLA Thu, 16 Apr 2026 10:43:05 +0000 DeepThinkVLA provides code, data/checkpoints, and LIBERO evals for a 2.9B pi0-FAST-derived hybrid VLA decoder that writes CoT reasoning before parallel action chunks. It reports 97.0% average LIBERO success, +15.5 points over naive autoregressive CoT, with Masked-CoT inference at 0.175x pi0-FAST autoregressive latency. — Tags: #RL #VLA #Other — ★ 523 — Score 35.639 https://github.com/NVIDIA/warp github:NVIDIA/warp Tue, 26 May 2026 21:44:11 +0000 NVIDIA Warp is a Python framework for writing GPU-accelerated simulation code aimed at robotics, machine learning, and physically based modeling. It is useful when researchers need high-performance differentiable or parallel simulation kernels without dropping fully into low-level CUDA, especially for workflows that mix simulation with learning systems. — Tags: #Other — ★ 6689 — Score 35.553 https://github.com/vla-safe/SAFE github:vla-safe/SAFE Thu, 21 May 2026 21:21:34 +0000 SAFE is the official codebase for a NeurIPS 2025 project on multitask failure detection in vision-language-action robot models. It focuses on detecting when VLA policies are likely to fail across tasks, which is useful for evaluating and adding safety checks around embodied AI systems before or during deployment. — Tags: #VLA #Other — ★ 71 — Score 35.5 https://github.com/RobotControlStack/robot-control-stack github:RobotControlStack/robot-control-stack Fri, 22 May 2026 20:07:54 +0000 RobotControlStack is a lightweight sim-to-real stack for training and deploying vision-language-action models and reinforcement-learning agents without depending on ROS. It provides native MuJoCo/Gymnasium wrappers with synchronous execution across common robot platforms including Franka, UR5e, xArm, and SO101, making it useful for researchers who want a leaner control and simulation pipeline for VLA or RL experiments. — Tags: #RL #VLA #Manipulator — ★ 98 — Score 34.954 https://github.com/ginwind/VLA-JEPA github:ginwind/VLA-JEPA Sat, 02 May 2026 03:55:52 +0000 VLA-JEPA provides training/evaluation code for augmenting a Qwen3-VL-2B VLA policy with a V-JEPA2 latent world/video model, built on starVLA. It supports LeRobot v2.1 robot data plus human-video training and includes LIBERO, LIBERO-Plus, and SimplerEnv eval scripts; no results are given. — Tags: #RL #VLA #Other — ★ 245 — Score 34.716 https://github.com/isaac-sim/Sim-to-Real-SO-101-Workshop github:isaac-sim/Sim-to-Real-SO-101-Workshop Wed, 13 May 2026 18:05:24 +0000 Train an SO-101 Robot From Sim-to-Real With NVIDIA Isaac Welcome to this workshop on sim-to-real transfer for the SO-101 robot! This repository contains the assets and code to accompany this learning content. The rest of this README will help you setup the environment and ensure everything is installed correctly. — Tags: #Other — ★ 38 — Score 34.704 https://github.com/shihao1895/MemoryVLA github:shihao1895/MemoryVLA Mon, 27 Apr 2026 03:52:31 +0000 MemoryVLA adds a hippocampal-like perceptual-cognitive memory module to VLA policies, targeting long-horizon manipulation from third-person RGB and language only. The repo releases OpenVLA-based MemoryVLA and Dexbotic-based MemoryVLA+, with checkpoints/logs; reported averages include 97.1 on LIBERO for MemoryVLA+ mix and 84.4 on Bridge. — Tags: #RL #VLA #Manipulator — ★ 244 — Score 34.54 https://github.com/unitreerobotics/unitree_rl_mjlab github:unitreerobotics/unitree_rl_mjlab Mon, 13 Apr 2026 13:44:11 +0000 Unitree RL Mjlab ✳️ Overview Unitree RL Mjlab is a reinforcement learning project built upon the mjlab, using MuJoCo as its physics simulation backend, currently supporting Unitree Go2, A2, As2, G1, R1, H1 2 and H2. — Tags: #RL #Other — ★ 375 — Score 34.425 https://github.com/IRMVLab/awesome-robot-learning-from-human-videos github:IRMVLab/awesome-robot-learning-from-human-videos Fri, 22 May 2026 09:56:46 +0000 Paper list for robot learning from human videos (LfHV) — Tags: #VLA #Other #Survey — ★ 110 — Score 34.31 https://github.com/MilkClouds/awesome-vla-study github:MilkClouds/awesome-vla-study Sat, 21 Mar 2026 03:45:32 +0000 Curated VLA reading syllabus organizes papers in a 14-week order from diffusion/flow-matching foundations through RT-1/RT-2, Octo/OpenVLA, current VLAs, data scaling, efficient inference, RL fine-tuning, reasoning, and world models. It is a study guide, not a new model, dataset, or benchmark. — Tags: #RL #VLA #Other #Survey — ★ 250 — Score 34.154 https://github.com/PKU-Alignment/VLA-Arena github:PKU-Alignment/VLA-Arena Sat, 14 Mar 2026 07:05:28 +0000 🤖 VLA-Arena: An Open-Source Framework for Benchmarking Vision-Language-Action Models VLA-Arena is an open-source benchmark for systematic evaluation of Vision-Language-Action (VLA) models. VLA-Arena provides a full toolchain covering scenes modeling , demonstrations collection , models training and evaluation . — Tags: #VLA #Other — ★ 165 — Score 34.028 https://github.com/keon/awesome-physical-ai github:keon/awesome-physical-ai Mon, 30 Mar 2026 22:14:05 +0000 Awesome Physical AI A curated list of academic papers and resources on Physical AI — focusing on Vision-Language-Action (VLA) models, world models, embodied ai, and robotic foundation models. — Tags: #RL #VLA #Other #Survey — ★ 253 — Score 34.016 https://github.com/PKU-Alignment/SafeVLA github:PKU-Alignment/SafeVLA Tue, 31 Mar 2026 20:53:14 +0000 SafeVLA is a GitHub repo for safety-aligning vision-language-action models using constrained learning. It is tied to a NeurIPS 2025 Spotlight paper, giving researchers a source entry point to inspect the method. — Tags: #VLA #Other — ★ 139 — Score 33.974