Small objects. Precise moves.
Transparent stationery, reflective coffee capsules and soft tea bags pose different handling challenges. From sorting and grasping to opening cases and stamping cards, @ace_robotics’ robots are developing fine-manipulation skills through real-world feedback.
These tasks form part of the retail training underway with Kuaikeda at the companies’ joint Deqing base.
#ACERobotics #Kuaikeda #EmbodiedAI #RoboticManipulation
33
A packed order. An organized shelf. A series of decisions behind each.
Robots must recognize products, sequence their actions and place items accurately. @ace_robotics and Kuaikeda are training these capabilities through order packing and shelf organization tasks, using embodied RSI and shared experience across robots to refine reusable retail skills.
#ACERobotics #Kuaikeda #EmbodiedAI #RetailRobotics
27
Handle delicate food. Follow the right sequence. Build a complete meal-preparation workflow.
From picking macarons and handling oden to assembling sandwiches and reheating buns, ACE Robotics is training robots for multi-step food preparation. Its dual-brain architecture and embodied recursive self-improvement (RSI) support continuous skill refinement through real-world feedback.
Part of @ace_robotics' collaboration with Kuaikeda at their joint retail skills training base in Deqing, Zhejiang.
#ACERobotics #Kuaikeda #EmbodiedAI #RetailRobotics
Food preparation. Order packing. Shelf organization. Fine manipulation.
ACE Robotics and Kuaikeda are training robots across everyday retail tasks at their joint retail skills training base in Deqing, Zhejiang—with the goal of deploying thousands of retail robots in 2027.
@ace_robotics's dual-brain architecture pairs a general-purpose model with the Kairos World Model. Embodied recursive self-improvement (RSI) uses real-world execution feedback to refine skills, while multi-robot learning enables experience sharing and skill reuse.
From training to store operations, the collaboration is building a reusable skill foundation for retail robotics at scale.
#ACERobotics #Kuaikeda #EmbodiedAI #RetailRobotics #InstantRetail
57
On a real-world plug-insertion task, general-purpose models explored different approaches and adjusted their strategy using collision detection and force analysis. ACE-CaP-0 distilled the successful method into a reusable robot skill, enabling the robot to complete the task in just one minute.
From an hour of trial and error to a skill that runs in a minute.
Explore ACE-CaP-0:
Report:lnkd.in/gGnSTQu8
Project:lnkd.in/gYsmEESg
#ACERobotics #ACECaP0 #EmbodiedAI #Robotics #RobotLearning
From hours of exploration to minutes of execution.
ACE-CaP-0 enables general-purpose models to gather experience, develop and refine robot skills, and turn proven strategies into reusable code. Across eight real-world manipulation tasks, it achieved a 92.1% success rate in 240 independent trials.
Explore ACE-CaP-0:
Report:acerobotics2025.github.io/AC…
Project:acerobotics2025.github.io/AC…
#ACERobotics #ACECaP0 #EmbodiedAI #Robotics #RobotLearning #PhysicalAI #AI
The 9th CIIE (@ciieonline) welcomed a new kind of security patrol. Powered by ACE ROBOTICS’ Embodied Super Brain Module A1, robotic dogs navigated busy indoor and outdoor spaces with the autonomy to perceive, decide and adapt in real time.
From dynamic route planning to AI-enabled incident alerts, they brought embodied intelligence into a demanding, real-world security operation, working alongside on-site teams to help keep the Expo moving safely.
#ACEROBOTICS #CIIE #EmbodiedAI #AutonomousRobotics #SmartSecurity
75
ACE Robotics brings its dual-brain architecture—General Model + Kairos World Model—to a real hotel laundry workflow. Mundo creates a physics-enabled digital twin for robot training, while embodied RSI turns execution feedback into continuously improving skills.
From digital training to real-world service, robots move beyond single actions toward end-to-end tasks.
#ACERobotics #WorldModel #EmbodiedAI #HospitalityRobotics #DigitalTwin #RSI
ACE ROBOTICS and @Siemens have signed a partnership agreement under the Siemens Xcelerator Ecosystem during the 9th China International Import Expo.
By connecting @ace_robotics’s expertise in world models and robots’ autonomous perception and decision-making with Siemens Xcelerator’s open ecosystem, the collaboration aims to bring technology closer to customer needs and advance embodied AI applications in industrial inspection and other scenarios.
#ACERobotics #Siemens #Robotics #EmbodiedAI #WorldModels @siemenssoftware
1
91
One environment. Navigation, manipulation, and 4D simulation.
Mundo’s physics-aware 3D proxy provides the spatial and semantic information needed for robot navigation, including objects, obstacles, and traversable areas. Different robots can attempt manipulation tasks in the same reconstructed environment, with external, head-mounted, and wrist-camera views generated alongside task trajectories and interaction states.
Mundo also supports 4D simulation: freeze an event at a chosen moment, move the camera, and inspect actions and occlusions from new viewpoints.
Project: mundo-world-model.github.io/
Blog: huggingface.co/blog/Luo-Yiha…
#ACEROBOTICS #Mundo #WorldModel #RobotSimulation #RobotManipulation #RealToSim
2
136
A world model should respond to actions.
Mundo represents objects with explicit names, poses, sizes, and physical properties such as mass and friction. Actions and applied forces are specified in code, while a physics engine computes motion, contact, and collisions under gravity and contact constraints.
Push an object, release it, or change its material and gravity. Mundo simulates the response and renders the resulting motion within the original environment, with velocity, contact points, and impact forces shown alongside.
Project: mundo-world-model.github.io/
Blog: huggingface.co/blog/Luo-Yiha…
#ACEROBOTICS #Mundo #WorldModel #PhysicsSimulation #EmbodiedAI
2
149
Explore a real place from just a few images.
Mundo turns 1–12 context images into an explorable 3D world, generating new views along specified camera paths while preserving the scene’s layout, appearance, and visible details. A single image or painting can also become the starting point for a new space.
With more context views, Mundo gains stronger visual grounding. For large environments, it streams generation along the same 3D proxy to create continuous camera paths lasting from tens of seconds to minutes.
Project: mundo-world-model.github.io/
Blog: huggingface.co/blog/Luo-Yiha…
#ACEROBOTICS #Mundo #WorldModel #SpatialExploration #GenerativeAI
1
115
Introducing Mundo, a customized world model developed by @ace_robotics with @MMLabNTU.
From just a few images of a specific environment, Mundo builds an explorable, editable, and simulatable 3D world. Its 3D proxy captures scene layout and objects; a code-based scene program defines their properties and actions; a physics engine computes motion, contact, and collisions; and a video model renders the results with the original visual context.
Mundo connects spatial exploration, physical interaction, robot navigation, manipulation, and 4D simulation in one environment, bringing real-world scenes into Real-to-Sim testing and robot training.
Project: mundo-world-model.github.io/
Blog: huggingface.co/blog/Luo-Yiha…
#ACEROBOTICS #Mundo #WorldModel #EmbodiedAI #Robotics #RealToSim
3
94
Spatial intelligence, now within reach.
@ace_robotics introduces the ACE-Brain-0.6 API—bringing spatial grounding capabilities to developers worldwide.
With training data scaled from 1.1 million to approximately 1.5 million samples, ACE-Brain-0.6 achieves an average score of 75.41 across seven public grounding benchmarks, extending its lead over GPT-6 Astra while also surpassing it in inference speed.
For robots, knowing what to do is only the beginning. Understanding where to act—and how to interact with the physical environment—is essential.
Through ACE Hub, developers can access ACE-Brain-0.6 without deploying the model themselves, exploring applications in robotic manipulation, smart retail, logistics sorting, and beyond.
Build with spatial intelligence. Bring your next idea into the physical world.
Explore the API: acehub.acerobotics.com/login
#ACERobotics #ACEBrain #SpatialIntelligence #EmbodiedAI #Robotics
ACE Robotics’ W1 fulfillment robot has received the 2026 @Forbes China Humanoid Robot Future Award.
Designed for demanding retail environments, W1 navigates aisles as narrow as 75 cm and handles both flexible packages and rigid goods with precision.
We’re building embodied AI that works in the real world.
#ACERobotics #W1 #EmbodiedAI #Robotics #ForbesChina @Forbes_china
1
48
At the 2026 IDC China CIO Summit in Shanghai, ACE Robotics received three honors: the Excellence Award, the Future Industry Star award for Kairos World Model, and recognition of Chairman Xiaogang Wang as a Leading Scientist in Digital Intelligence.
Kairos 3.1 brings together world understanding, generation, and prediction to advance embodied AI from models to real-world applications.
#ACERobotics #EmbodiedAI #WorldModel #IDC @idcchina
1
54
ACE-Brain-0.6 is here.
A 9B embodied model, trained on nearly one million verifiable spatial grounding samples, outperforms GPT-6 Astra on six grounding benchmarks.
· Nearly 1 million verifiable spatial grounding samples
· 74.53 average across 7 public benchmarks
· 94.13 on RobotGrounding
From seeing the world to locating the right target.
#ACEROBOTICS #ACEBrain06 #EmbodiedAI #SpatialIntelligence #VisualGrounding #Robotics #PhysicalIntelligence #AIModels
5
96
ACE ROBOTICS has officially released ACE-Brain-0.6, a vision-space grounding enhanced embodied foundation model.
Built on Qwen3.5-9B and trained with nearly one million high-quality, verifiable spatial grounding samples, ACE-Brain-0.6 achieves:
🌟6 out of 8 Grounding benchmarks ahead of OpenAI GPT-6 Astra
🌟74.53 average score across 7 public benchmarks, vs. 73.68 for GPT-6 Astra
🌟94.13 on RobotGrounding, vs. 91.10 for GPT-6 Astra
The model targets object localization, spatial relationships, affordance understanding, interaction-region prediction, and reliable target absence detection.
From understanding the world to locating the right target, ACE-Brain-0.6 explores a data-scaling path toward more precise and dependable physical intelligence.
Learn More:
📄ArXiv:arxiv.org/abs/2610.27040
🌐Project:daxiaorobotics.github.io/ACE…
💻GitHub:GitHub - DAXIAORobotics/ACE-Brain-0.6: The official repository of ACE-Brain-0.6 physical foundation
🤗Hugging Face:huggingface.co/ACERobotics/A…
#ACEROBOTICS #ACEBrain06 #EmbodiedAI #SpatialIntelligence #VisualGrounding #Robotics #PhysicalIntelligence #AIModels
1
55














