AB
AiBoss
project

White Tiger-VTouch - An open-source cross-ontology visual-tactile multimodal dataset from the National Geographic Center.

White Tiger-VTouch is the world's first and largest cross-body visual-tactile multimodal dataset, open-sourced by the National and Local Jointly Constructed Humanoid Robotics Innovation Center in collaboration with VTouch Robotics. The dataset includes visual-tactile sensor data, RGB-D data, and more...

What is White Tiger-VTouch?

Baihu-VTouch is the world's first and largest cross-body visual-tactile multimodal dataset, open-sourced by the National and Local Jointly Constructed Humanoid Robot Innovation Center in collaboration with VTouch Robotics. The dataset includes visual-tactile sensor data, RGB-D data, joint pose data, and more, covering various robot body configurations, with a data scale exceeding 60,000 minutes. The dataset pioneers a cross-body real-interaction acquisition paradigm, overcoming core bottlenecks in embodied intelligent robots such as data scarcity, insufficient tactile information, and weak generalization ability, providing key corpora and engineering foundations for building embodied basic models.

Main functions of White Tiger-VTouch

  • Provides real physical interaction dataThe dataset fills the gap in large-scale real-world visual-tactile interaction data, providing a data foundation for the physical operation and fine manipulation of robots in complex environments.
  • Supports cross-ontology generalization capabilitiesBy acquiring cross-entity data, it supports model training for different robot configurations (such as wheeled robots, bipedal robots, etc.) and enhances the generalization ability of the model.
  • Assisting in the development of embodied intelligent modelsThis provides key corpus for building embodied basic models with physical understanding and fine manipulation capabilities, accelerating the leap of robots from "being able to see" to "being able to touch and control".
  • Promote standardizationAs a landmark achievement of the national standardization pilot project in the field of embodied intelligence, it helps to build and implement the national pilot standard system for embodied intelligence training grounds.

The technical principles of White Tiger-VTouch

  • Multimodal data acquisitionIt combines multiple sensors such as visual and tactile sensors, RGB-D cameras, and joint pose sensors to collect multimodal data including vision, touch, and force.
  • Real physical interactionBy collecting data from real physical interaction scenarios, it records the pressure distribution, deformation data, etc. during the contact process of objects, and provides high-fidelity tactile information.
  • "Matrix-style" task constructionBreaking away from the traditional single-task data collection model, we designed a "matrix" task structure that covers multiple scenarios and task types, enabling large-scale data generation and structured capability coverage.
  • Cross-level interactive understanding annotation: Construct a cross-level interactive understanding embodied annotation system, and achieve cross-modal representation learning and unified understanding through joint modeling and alignment of multimodal semantics such as vision, language, action and touch.
  • Unified Algorithm Framework: Construct a unified training-inference algorithm framework for real robot deployment, realize a complete closed loop from multimodal data processing to model training and online inference, and improve the efficiency of model development and verification.

White Tiger-VTouch Project Address

  • OpenLoong open source communityhttps://ai.atomgit.com/openloong/visuo-tactile

Application scenarios of White Tiger-VTouch

  • Robot Operation and InteractionIt provides high-precision visual and tactile data support for robots in scenarios such as home services, industrial manufacturing, and special operations, helping robots complete complex operational tasks.
  • Artificial intelligence and machine learningThe dataset provides large-scale multimodal data for training embodied intelligence models, reinforcement learning, and cross-modal learning, enhancing robots' learning and interaction capabilities in the real world.
  • Smart terminals and human-computer interactionIt supports the development of more natural human-computer interaction methods for handheld smart devices and smart wearable devices, enhancing virtual reality and augmented reality experiences.
  • Education and ResearchIt provides abundant research resources for universities and research institutions, supporting cutting-edge research and educational curriculum development in fields such as robotics and artificial intelligence.
  • Medical treatment and rehabilitationThis technology enables rehabilitation robots and surgical assistive robots to achieve more precise operations through visual and tactile data, thereby improving the effectiveness of medical rehabilitation and the safety of surgery.