Beijing Issues Action Plan to Boost Embodied Intelligence Foundation Model Capabilities

北京：大幅提升具身智能基础模型能力

Summary

According to Beijing's Economic and Information Commission, the city has issued the 'Action Plan for High-Quality Development of the Embodied Intelligence Robot Industry (2026–2028)'. The plan aims to significantly enhance embodied intelligence foundation model capabilities by supporting multiple technical routes including visual-language-action models, world models, and reinforcement learning. Key focus areas include multi-source information fusion, physical space understanding, human emotion recognition, real-time reasoning, cross-embodiment transfer, self-learning evolution, group collaboration, long-term memory, lightweight local deployment, and model interpretability. The goal is to achieve an international first-class level in embodied intelligence foundation models.

Source: Chinese robotics — Robot learning research · Read original article ↗

Article text · Machine translation into English

Beijing: Significantly enhance embodied intelligence foundation model capabilities

First Finance 2026-11-11 12:43:49

Editor: Zhang Xiao

Beijing: Significantly enhance embodied intelligence foundation model capabilities

According to Beijing Economic and Information News, the “Beijing Embodied AI Robot Industry High-Quality Development Action Plan (2026–2028)” has been issued. The plan proposes significantly improving the capabilities of embodied AI foundation models. It supports the layout and technological fusion development of multiple technical routes such as Vision-Language-Action models (VLA), World Models (WM), and reinforcement learning, leading technological innovation in all major routes. It supports tackling technologies such as multi-source information fusion, physical space understanding, and human emotion recognition, pushing for a leap from “perception” to “cognition,” where “brain” cognition reaches the level of understanding physical law constraints and actively recognizing non-verbal instructions such as human actions and expressions; it supports tackling technologies such as real-time inference, cross-embodiment transfer, and self-learning evolution, significantly improving “cerebellum” capabilities for multi-task generalization, high-reliability execution, fast-paced operations, and ultra-precise manipulation; it supports tackling technologies such as swarm collaboration, long-term memory, localized lightweight deployment, and model explainability, significantly improving the real-world application deployment capabilities of embodied AI models, with the overall embodied AI foundation models reaching international first-class levels.

Report

What the source reports

Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.

paper: not_reported

Source excerpts and review record

Automatically extracted; no manual editorial approval recorded.