Enabling Humanoid Robots With Human-Level Dexterous Manipulation

CORE CAPABILITIES

From perception to execution - one interface

深度理解

Perception

Object detection, pose estimation, spatial and scene understanding
指缘地点6

Pick & Place

Grasp generation, stable placement, and outcome verification
手部修复

Dexterous Manipulation

Multi-finger coordination, in-hand adjustment, insertion, and tool use
运动控制卡

Tactile & Force Control

Contact sensing, slip detection, pressure feedback, and anomaly response
设置_触觉反馈

Motion & Control

Trajectory planning, bimanual coordination, and real-time execution
信号诊断

Status & Diagnostics

Unified management of device, skill, task, and error states

SKILL LIBRARY

Call a skill. Skip the rebuild

Object Detection

Identify and classify objects in the scene

Pose Estimation

Output position and orientation of a target object

Grasp Generation

Generate optimal grasp candidates for a given target

Pick & Place

Execute full grasp, transport, and precise placement

Object Reorientation

Adjust and reposition an object to a target pose

Precision Insertion

High-accuracy alignment and insertion operations

Button Pressing

Detect and actuate buttons and switch-type controls

Switch & Knob

Handle rotary, toggle, and slider control interactions

Tool Manipulation

Identify, grasp, and operate task-specific tools

Deformable Handling

Manipulate flexible objects such as fabric or cables

Bimanual Manipulation

Coordinate dual-arm operations for two-hand tasks

Human-Robot Handover

Safely exchange objects with a human operator

Tactile Grasping

Adaptive grasp control using force and tactile feedback

Skill Status Query

Retrieve runtime state and progress of active skills

Abort & Recover

Handle exceptions and safely resume task execution
 

MODEL & AGENT INTEGRATION
Open to the full spectrum of robot intelligence

 
Robot Foundation Models
General representations, cross-task transfer, post-training and fine-tuning
Vision-Language-Action Models
Map visual observations and language instructions to actions or task sequences
VLMs & Multimodal LLMs
Scene understanding, object grounding, task decomposition, and constraint generation
Imitation Learning & Policies
Behavior cloning, diffusion policies, Transformer policies, and multi-task strategies
Reinforcement Learning & World Models
Policy optimization, long-horizon planning, sim training, and sim-to-real transfer
Motion Planning & Control Models
Grasp planning, trajectory optimization, MPC, and whole-body control
Multi-Agent & Task Planners
Skill selection, tool calling, task orchestration, memory, and error recovery
Custom Enterprise Models
Connect proprietary, domain-specific, or specialized control models via adapters

COMPOSITE TASKS

Multiple skills. One complete execution

Smart Sorting

Detect  Grasp  Classify  Place  Confirm

Precision Assembly

Locate  Adjust  Align  Insert  Force-verify

Smart Sorting

Select  Bimanual grasp  Transfer  Pack  Review

Precision Assembly

Assess shape  Unfold  Align  Arrange  Confirm

Smart Sorting

Identify tool  Grasp  Reorient  Operate  Return

Precision Assembly

Read intent  Confirm task  Execute  Hand off  Feedback
 

DATA FLYWHEEL
Every execution improves the next

 
61多平台硬件适配
  Log task context, skill versions, 
         and robot configurations in a unified forma
  Pinpoint failure stages and 
          capture high-value real-world robot data
  Support model post-training, policy optimization, 
          and cross-embodiment adaptation
  Close the loop: 
          collect    train    deploy    evaluate    retrain

PLATFORM & DEPLOYMENT

One capability set. Any hardware. Any environment

On-Premise
Low latency, stable operation, and data privacy
Edge

Deploy perception, policy, and execution on the robot or edge compute platform

Cloud
Centralized management of models, skills, tasks, devices, and data
Hybrid
Real-time control on-device; training, analytics, and operations in the cloud