x-square-robot/X-Planner-9B-0916
X-Planner-9B-0916 is a 9.4 billion parameter multimodal planner developed by X-Square-Robot, based on the Qwen3.5 architecture. Designed for long-horizon robot manipulation, it processes high-level instructions and multi-view observations to predict action-grounded events. This model excels at generating structured task plans for embodied intelligence, offering both readable event mode and compact latent planning states.
Loading preview...
X-Planner-9B-0916: Event-Structured Task Planning
X-Planner-9B-0916 is a 9.4 billion parameter multimodal planner developed by X-Square-Robot, specifically designed for long-horizon robot manipulation. Built upon the Qwen3.5 architecture, this model acts as a planning front-end, taking high-level instructions, synchronized multi-view observations, and optional execution history to predict the next action-grounded event. The resulting structured plan is then passed to a downstream world-action model, making the intermediate planning state explicit and inspectable.
Key Capabilities
- Event-Structured Planning: Generates structured natural-language planning states, making robot task execution transparent and interpretable.
- Unified Latent Planning: Supports a unified mode that produces compact latent planning states using Staircase Decoding for efficient downstream processing.
- Multimodal Input: Integrates multi-view observations alongside textual instructions for comprehensive environmental understanding.
- Qwen3.5 Architecture: Leverages the robust Qwen3.5 9B architecture for its planning capabilities.
Good For
- Embodied Intelligence Research: Ideal for researchers and developers working on complex robotic manipulation tasks requiring explicit and inspectable planning.
- Robot Task Sequencing: Useful for generating detailed, step-by-step plans for robots to execute long-horizon tasks.
- Developing Downstream World-Action Models: Provides a structured planning output that can seamlessly integrate with and guide subsequent world-action models.