Embodied-R1.5: Evolving Physical Intelligence via Embodied Foundation Models

2026-06-09 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

Embodied-R1.5 is introduced as a unified Embodied Foundation Model (EFM) designed to integrate comprehensive embodied reasoning, including cognition, task planning, correction, and pointing, within a single architecture for general physical intelligence. The model leverages three automated data construction pipelines to build a large-scale data system of over 15B tokens and employs a multi-task balanced RL recipe to manage heterogeneous task conflicts. It also features a Planner-Grounder-Corrector (PGC) closed-loop framework, enabling autonomous execution and self-correction for long-horizon tasks. With only 8B parameters, Embodied-R1.5 achieves state-of-the-art performance on 16 out of 24 embodied VLM benchmarks, outperforming models like Gemini-Robotics-ER-1.5 and GPT-5.4. It can be fine-tuned into a Visual Language Agent (VLA) with minimal data, surpassing $π_{0.5}$ across four popular manipulation benchmark suites. Extensive zero-shot real-robot experiments confirm its strong generalization in instruction following and complex manipulation tasks. The project open-sources model weights, datasets, training code, and EmbodiedEvalKit.

Key takeaway

For robotics engineers developing embodied AI systems, Embodied-R1.5 offers a powerful foundation for general physical intelligence. You should consider integrating this 8B-parameter EFM, especially given its state-of-the-art performance across 16 VLM benchmarks and strong real-robot generalization. Utilize the open-sourced model weights and EmbodiedEvalKit to accelerate your development and evaluation of long-horizon, self-correcting robotic tasks.

Key insights

Embodied-R1.5 unifies diverse embodied reasoning capabilities into an 8B-parameter EFM, achieving SOTA performance and real-world generalization.

Principles

• Unified architecture for diverse embodied reasoning.
• Large-scale automated data construction improves coverage.
• Closed-loop PGC framework enables self-correction.

Method

The Planner-Grounder-Corrector (PGC) framework enables autonomous execution and self-correction by integrating planning, grounding, and corrective actions in a closed loop.

In practice

• Fine-tune Embodied-R1.5 into a VLA with small datasets.
• Use EmbodiedEvalKit for embodied task evaluation.
• Apply zero-shot for real-robot instruction following.

Topics

• Embodied Foundation Models
• Robotics
• Visual Language Models
• Embodied AI
• Planner-Grounder-Corrector
• Open-source AI

Best for: Research Scientist, AI Scientist, Robotics Engineer, Machine Learning Engineer

Related on AIssential

• Sergey Levine - Building LLMs for the Physical World - [Invest Like the Best, EP.465] — The core idea is building general robotic foundation models for universal control of embodied systems.
• A benchmarking framework for embodied neuromorphic agents — A new framework standardizes benchmarking for neuromorphic control of soft robots in dynamic, real-world environments.
• The Sequence Knowledge #842: Everything You Need to Know About World Models — World models represent the shift from text-based AI to physical simulation, enabling AI to understand and operate within dynamic reality.
• Humanoid Robots Exit Labs: Mapping the Technical Path to Embodied AI at AW 2026 — Humanoid robots are evolving into embodied AI platforms, transitioning from labs to industrial applications via data-driven iteration and distributed computing.
• A peek inside Physical Intelligence, the startup building Silicon Valley’s buzziest robot brains — Physical Intelligence aims to build general-purpose robotic intelligence through extensive data collection and cross-embodiment learning.
• A robot operating system framework for using large language models in embodied AI — Connecting LLM agents to ROS enables versatile, robust, and scalable embodied AI through natural language.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.