Tactile Embodied Foundation Models

EN
Tactile Embodied Foundation Models
Home Products Embodied Foundation Models Tactile Embodied Foundation Models
Product Overview

The company focuses on the R&D of a new generation of tactile-enabled embodied large models, deeply integrating tactile information into its core technical roadmap.

Unlike conventional embodied models that rely on vision alone, we fuse vision, touch and force end to end, breaking the bottleneck of robots that can only see but cannot feel. This enables the leap from passively executing commands to actively understanding the world, giving robots autonomous cognition of the physical world, reasoning about object properties and interaction logic, and autonomous task planning and execution in complex dynamic scenes.

The model will provide new intelligent decision-making solutions for industrial precision assembly, compliant manipulation by service robots and the automation of scientific experiments.

Product Advantages

VTLA Model · Multimodal fusion
A multimodal foundation model that fuses visual and tactile information, breaking the limits of a single sense and understanding environments and task instructions more comprehensively.

Tactile World Model · Physical understanding
A physical-world model trained on large-scale tactile datasets that significantly improves a robot's understanding of physical properties, prediction of future states and motion planning.

Tactile Reinforcement Learning · Autonomous Learning
Reinforcement learning driven by tactile feedback lets robots optimize manipulation policies through repeated trial, error and interaction, achieving genuine autonomous learning and evolution.