Contact sales
Demo videos

Give robots a
sense of touch.

E-skin sensors that turn physical contact into precise tactile data.

Explore our products

Tactile intelligence

Bring human-level dexterity to robots through tactile intelligence.

Carus AI is a full-stack platform for embodied AI dexterous skills. Built on in-house tactile sensor systems, we close the loop from dexterous data-collection hardware and dexterous data to dexterous skills and real-world deployment.

  1. 01Tactile & Sensing
  2. 02Robot Hand & Arm
  3. 03Dexterous Data Infrastructure
  4. 04Skills API

Demo videos

From human touch
to robot skill.

A human hand in a tactile glove and a robot hand each grasping a white cube
A white robotic hand gently grasping a white cylinder
Two robotic hands folding a blue cloth
The Flex Hand F25 dexterous robot hand writing with a pen
An operator wearing a head-mounted camera collecting manipulation data next to a robot arm
01 / 05

Human demonstration, robot execution

Dual-source data: people show it once, robots learn to repeat it.

Sensor

Every contact
becomes data.

Flexible tactile e-skin for robot hands and gloves. It maps where contact happens, how hard, and how it changes over time — the signal robots need for adaptive, force-aware grasping.

  • FPS-01 Tactile E-SkinFlexible
  • Hybrid E-SkinHybrid architecture
  • Flexible Tactile E-SkinRobot hands
  • Fabric Tactile Sensors & GlovesTextile
See tactile products

Simulation · Pinch

Grip force0.00 N
Active taxels0
Fingers in contact0 / 5

Press a fingertip

Products

Built for dexterity.

Sensors, hands and capture hardware, designed to work together.

Data

From human hands
to robot skills.

A closed loop from dexterous data collection to deployable skills, bridging human demonstration and robot execution.

  1. An operator collecting synchronized vision, motion and tactile data
    01 — Capture

    Vision-tactile-motion hardware

    Glove body, collection backpack and head vision system record how people really manipulate objects.

  2. Carus data processing platform
    02 — Structure

    Multimodal dexterous database

    Multi-view vision, hand kinematics and robot joint state, tactile and force sensing, task semantics and temporal annotation.

  3. A grid of real-world manipulation tasks
    03 — Deploy

    Skills for real-world tasks

    Pick and place, in-hand manipulation, precision assembly, tool use and bimanual work, served through one API.

Type A

Human demonstration data

Embodiment-agnostic. Accumulates fine manipulation priors at scale and transfers across platforms.

Type B

Robot execution data

Teleoperated platforms capture the full perception → decision → execution pipeline, aligned to specific action spaces.

Dual-source synergy · pre-training · fine-tuning · evaluation · cross-embodiment transfer

  • Standardized datasets

    Cleaned, annotated and structured for model training and evaluation.

  • Custom collection

    Targeted data collection for specified tasks, scenes and robot platforms.

  • Training data engineering

    Action conversion, episode curation and training-format adaptation.

  • Data governance

    Quality assessment, version control and private data management.

API

Call a skill.
Skip the rebuild.

From perception to execution, one interface — open to foundation models, VLAs, policies and planners.

  • 01Object detection
  • 02Pose estimation
  • 03Grasp generation
  • 04Pick & place
  • 05Object reorientation
  • 06Precision insertion
  • 07Button pressing
  • 08Switch & knob
  • 09Tool manipulation
  • 10Deformable handling
  • 11Bimanual manipulation
  • 12Human-robot handover
  • 13Tactile grasping
  • 14Skill status query
  • 15Abort & recover

Composite tasks — multiple skills, one execution

Smart sorting

  1. Detect
  2. Grasp
  3. Classify
  4. Place
  5. Confirm

Precision assembly

  1. Locate
  2. Adjust
  3. Align
  4. Insert
  5. Force-verify

Bin packing

  1. Select
  2. Bimanual grasp
  3. Transfer
  4. Pack
  5. Review

Model & agent integration

  • Robot foundation models
  • Vision-language-action models
  • VLMs & multimodal LLMs
  • Imitation learning & policies
  • Reinforcement learning & world models
  • Motion planning & control
  • Multi-agent & task planners
  • Custom enterprise models

About

Our vision: build the world’s most valuable dexterous data platform for Physical AI.

Team

A world-class international team spanning technology R&D, industrialization and commercial expansion.

Headquarters

San Jose, California

Contact

Put a sense of touch in every robot hand.

Tell us what your robots need to handle — sensors, hands, data or skills.

support@carusai.com