Embodied AI information hub
Open-source projects, models, and datasets — plus a verified deployment matrix. More objective information to help founders make decisions.
Verified deployments
First 5 user submissions this month pin here
RobotEra
L3
China Post Guangzhou feed/sort with M7
Inbound feeding on China Post express lines at Guangzhou regional mail center (Jianggao): the RobotEra M7 locates parcels by position, size, shape and shipping label, grasps and flips them face-up, places them on the conveyor, and flags anomalies. Soft packs and rigid boxes of mixed sizes are handled in cadence with site automation (arms, unmanned forklifts). Peak feed rate is reported at about 1,200 parcels/hour and ≥85% of manual station throughput. Shopfloor video (e.g. Bilibili BV1acun6vE8C) and multi-outlet May 2026 coverage describe live operations rather than a lab demo. Unit shipment counts and deploy-line revenue are not publicly broken out.
View details →
Agility Robotics
L3
Commercial RaaS tote handling at GXO
Takes totes from collaborative robots and places them on conveyors; multi-task warehouse handling under a commercial RaaS deployment.
View details →
AgiBot
L3
Genie G2 on Longcheer tablet line
Load/unload on tablet QC stations, synced to line takt; live factory verification streams.
View details →
Unitree
L3
G1 commercial pilots (cleaning / airport)
Commercial facility cleaning/maintenance and airport cargo/luggage pilots among other scenes.
View details →
UBTECH
L3
Scaled Walker deploy at OEM plants
Tote handling, parts sorting, inspection, flexible material handling, and multi-robot collaboration on automotive lines.
View details →
Galbot
L2
Yibin ~CNY 236M / 500-unit public procurement
Public procurement for the Yibin high-speed rail south district: Beijing Galbot ranked first bidder at about CNY 235.92M for 500 embodied devices and supporting gear — 380 wheeled robots, 80 heavy-load robots, 10 quadrupeds, and 30 robot retail capsules. Media citing tender materials describe phased delivery over 12 months with a first batch of 250 units. This is a commercial bulk purchase (not a lab demo or Spring Festival stage show). Unit-level on-site commissioning status after award is not fully independently audited in open sources.
View details →
Figure AI
L1
BMW reports results of Figure 02 production pilot
Sheet-metal positioning for welding during a completed ten-month production pilot; not a claim of an ongoing fleet rollout.
View details →
Galbot
L1
Baida Precision 1,000+ framework (media)
In late December 2025, media report that Galbot and Baida Precision (百达精工) signed a strategic cooperation to deploy more than 1,000 Galbot embodied robots across Baida Precision and its ecosystem for complex precision-manufacturing scenes (raw material storage through machining and inspection narratives appear in secondary coverage). This is treated as a large industrial framework order / cooperation scale claim. Actual installed unit count per plant, SKU mix, unit price, and verified ROI are not broken out in independent primary filings found at audit.
View details →
Galbot
L1
S1 @ CATL line (media; KPIs disputed)
Multiple 2026 media pieces report that Galbot’s heavy-load embodied robot S1 has entered CATL (宁德时代) smart production lines for material handling / pick tasks on module and pack-related flows, with marketing claims of zero teleoperation and long-horizon continuous operation (including 7×24 and multi-month narratives in secondary articles). CATL battery partnership / aftermarket service framing also appears. Unit count, MTBF, success rate, and independent CATL press confirmation of the long-run KPIs are thin or disputed at audit — metrics kept as media-aligned claims only.
View details →
Hand-picked high-quality resources to anchor your map of the field
Fast and embedded solvers for nonlinear optimal control and MPC — used on real robots.
Why: Go-to open MPC / optimal-control stack for robots. Latest GitHub release: v0.6.0 — https://github.com/acados/acados/releases/tag/v0.6.0.
Long-standing robot simulator tightly integrated with ROS — sensors, multi-robot, and outdoor worlds.
Why: ROS-centric sim still widely used in industry labs. Latest GitHub release: gz-sim10_10.0.0 — https://github.com/gazebosim/gz-sim/releases/tag/gz-sim10_10.0.0.
Next-gen general physics engine emphasizing speed and usability for robots and generative scenes.
Why: Genesis targets high-throughput, Python-first GPU simulation for RL and data generation. Choose it when rollout count matters more than matching every classic MuJoCo tutorial. Expect a newer ecosystem and evolving APIs — read the official install notes for CUDA. Good for research teams already comfortable with parallel training. Latest GitHub release: v1.4.3 — https://github.com/Genesis-Embodied-AI/genesis-world/releases/tag/v1.4.3.
NVIDIA robot learning framework on Isaac Sim for large-scale parallel training and sim-to-real.
Why: Isaac Lab sits on the NVIDIA Isaac / Omniverse path for large-scale training and sensor-rich stacks. Pick it when you already plan NVIDIA GPUs and industrial-style pipelines. Install and concepts cost more than MuJoCo — not the first weekend toy. Best for labs moving toward production-minded sim training. Latest GitHub release: v3.0.0-EA — https://github.com/isaac-sim/IsaacLab/releases/tag/v3.0.0-EA.
Hugging Face open-source stack for embodied AI with unified data formats and training — great for imitation learning and VLA experiments.
Why: LeRobot is a hands-on hub for imitation and real-robot recipes with approachable docs. Recommended as a first from-dataset-to-policy path on low-cost hardware. Beginner-friendly if you follow official guides; still expect hardware and camera setup work for real robots. Latest GitHub release: v0.6.1 — https://github.com/huggingface/lerobot/releases/tag/v0.6.1.
Classic physics engine widely used in robot RL and control research — lightweight and accurate.
Why: MuJoCo is the default physics stack for a huge share of robot-learning papers and courses. Install is relatively light, Menagerie gives ready robot models, and tutorials are everywhere. Best if you want fast iteration on locomotion or manipulation before committing to a heavy GPU sim. Intermediate once you care about contacts and actuators; beginner-friendly for reading code and running demos. Latest GitHub release: 3.15.0 — https://github.com/google-deepmind/mujoco/releases/tag/3.15.0.
Stanford BEHAVIOR / OmniGibson household embodied benchmark with long-horizon activities in rich scenes.
Why: Leading household long-horizon embodied benchmark
Browser-native WebGL 3D library — the lowest-friction way to visualize robots, scenes, and simple kinematics without installing a physics stack. Ideal before MuJoCo/Isaac.
Why: Lowest install barrier for 3D robot/scene intuition — run in the browser before heavy sim stacks
Newest additions and updates by update date
Open-source multibody dynamics and robotics toolkit with C++ and Python interfaces.
NVIDIA synthetic robot trajectories for cross-embodiment GR00T post-training; download task subsets from the official dataset card.
World's first open-source enterprise world model.
Fast and embedded solvers for nonlinear optimal control and MPC — used on real robots.
Long-standing robot simulator tightly integrated with ROS — sensors, multi-robot, and outdoor worlds.
Next-gen general physics engine emphasizing speed and usability for robots and generative scenes.
Missing a great resource?
Submit open-source projects, models, or datasets. We review and list them as soon as we can.
