Simple AI 深朴智能
"Make The World Simple"
Beijing embodied-AI startup building living-space service robots; its HiFi-UMI robot-free capture rig shipped with an open 2,000-hour dataset.
Simple AI is a Beijing embodied-AI company developing general-purpose robots for human living spaces together with the models and training data needed to run them. Its headquarters is on the 21st floor of Building 1, No. 8 Courtyard, Beichen East Road, Chaoyang District, with further offices in Hangzhou and Hong Kong. Founder and chief executive Xiaofei Li holds bachelor's and doctoral degrees from Tsinghua University and previously worked on an L4 autonomous-driving architecture; in June 2026 he was named to 36Kr's Under 36 list.
The company works in four layers: foundation models, high-fidelity data, robot hardware and deployment. Its robot platform, which has no public model name yet, is specified on the official site with 23 degrees of freedom across the whole body, 7-degree-of-freedom arms each carrying a 5 kg end payload, a 2,000 TOPS main compute chip, two LiDARs, two RGB-D cameras and six GMSL camera links, with a split between a cloud brain and on-device large and small brains. On the software side, Omni-Simple-World Model is a unified embodied model that combines task understanding, world prediction and action generation for long-horizon tasks, and a memory-augmented Agentic OS uses an agent harness with hierarchical memory. Target solutions are integrated robot services for hotels, homes, senior-care facilities and retail.
The best documented product is HiFi-UMI, released and open-sourced on 30 July 2026. It is a portable, robot-free data-production system: an operator wears a head unit with a stereo camera and inertial sensors and uses two handheld gripper modules, each with two non-parallel fisheye cameras giving about 200 degrees of coverage. A shared hardware trigger aligns all sensors to under 40 microseconds, offline stereo-inertial SLAM gives about 3 mm end-effector accuracy, and the relative pose between the two grippers is measured directly. Automatic trajectory reconstruction and simulation replay checks each pass about 98 percent of recordings. In the company's technical report (arXiv 2607.25895), policies trained only on HiFi-UMI data came within a few percentage points of teleoperation-trained policies across three model architectures and four bimanual tabletop task groups, although the comparison used roughly ten times more HiFi-UMI trajectories. Alongside the system the company published HiFi-UMI-2K on Hugging Face under CC BY 4.0: 2,000 hours, more than 482,000 episodes and over 110 scenes with multi-view video, bimanual trajectories, gripper states, language annotations and subtask boundaries, with faces masked. The full internal corpus is stated at over 20,000 hours and 4.32 million episodes across more than 480 scenes.
Simple AI says it completed four funding rounds within a year. A Pre-A round of several hundred million yuan announced in June 2026 was led by ride-hailing group Didi, with Plum Ventures and sensor maker Keli Sensing joining and earlier investors including Linear Capital and Puhua Capital adding to their stakes. A Pre-A+ round, also of several hundred million yuan, followed on 3 September 2026 with a group of industrial and government-linked funds and existing investor Linear Capital. No customers or deployments have been named publicly.
Primary type & automation activities this supplier delivers:
HiFi-UMI
- Model: HiFi-UMI
- Launch Year: 2026
Portable robot-free capture rig that records human bimanual demonstrations as training data for robot arms.
HiFi-UMI-2K dataset
- Model: HiFi-UMI-2K
- Launch Year: 2026
Open 2,000-hour bimanual manipulation dataset collected with HiFi-UMI, released under CC BY 4.0.
Omni-Simple-World Model
Unified embodied model combining task understanding, world prediction and action generation for long-horizon tasks.
More related news
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
-
9
IROS 2026 Best Paper Finalists: South Korea Leads RobotToday · Oct 07, 2026
-
10
PTC Onshape at IMTS 2026: AI That Writes Its Own CAD Tools RobotToday · Oct 02, 2026
-
11
Siemens on Robots, Agentic AI and the "Last 3%" Problem RobotToday · Sep 23, 2026
-
12
Stratasys at IMTS 2026: GM Scales Additive to 20+ Plants RobotToday · Sep 15, 2026
-
13
Ailoys Brings Acoustic Industrial AI to IMTS 2026 RobotToday · Sep 10, 2026
- 14
-
15
China’s Dexterous Hand Industry, 2026 Part 1 — The Demand Picture RobotToday · Sep 04, 2026
-
16
China’s Dexterous Hand Industry, 2026 Part 2 — The Technology RobotToday · Sep 04, 2026
-
17
ROBOTTODAY Weekly August 10 – 14, 2026 RobotToday · Aug 15, 2026
-
18
Underactuated Robotics Russ Tedrake — RSF Specialist Shelf · Book S3 of 6 RobotToday · Aug 01, 2026
-
19
Introduction to Robotics J. J. Craig — RSF Top 10 Robot Education Textbooks · No. 6 RobotToday · Jul 31, 2026
- 20
Contact Simple AI
WEBSITE
https://simpleai.techPHONE
+86 400 606 5301HEADQUARTERS
China
Company Facts
Founded
-
Primary Role
Software/Algorithm
Company Size
-
Primary Region
Asia-Pacific
Annual Sales
-
Funding Stage
Early-Stage VC
Funding Total
-
Keywords
- Simple AI
- 深朴智能
- Shenpu Intelligence
- Simple World Lab (research arm)
- robot-free data collection
- universal manipulation interface
- bimanual manipulation
- embodied foundation model
- world model
- open dataset
- vision-language-action
- household robot
- Software/Algorithm
- Manufacturer
- Commercial Service Robot
- Perception & Vision
- Control & Software
- Cloud & Data
- Education & Research
- Retail & Hospitality
