🔥36氪•Stalecollected in 12m
Digua Robot and Qianxun Intelligence Partner on VLA
#robotics#vla#edge-ai#embodied-aispirit-v1.5-/-sunray-s600digua-robotqianxun-intelligencespirit-v1.5sunray-s600
💡New high-performance VLA model and chip combo for embodied AI developers.
⚡ 30-Second TL;DR
What Changed
Spirit v1.5 is an open-source VLA model outperforming Pi0.5
Why It Matters
This partnership accelerates the deployment of embodied AI by bridging the gap between advanced VLA models and specialized edge hardware.
What To Do Next
Evaluate the Spirit v1.5 model for your robotics projects if you require high-performance edge inference capabilities.
Who should care:Developers & AI Engineers
Key Points
- •Spirit v1.5 is an open-source VLA model outperforming Pi0.5
- •Sunray S600 chip delivers 560TOPS (INT8) AI computing power
- •Focus on high-performance edge inference for robotics
🧠 Deep Insight
Web-grounded analysis with 11 cited sources.
🔑 Enhanced Key Takeaways
- •Spirit v1.5, developed by embodied AI startup Spirit AI, achieved the top spot on the RoboChallenge Table30 leaderboard, a global real-hardware benchmark for embodied intelligence models, with a composite score of 66.09 and a 50.33% success rate, surpassing the previously dominant Pi0.5 model.
- •Spirit AI has open-sourced the Spirit v1.5 VLA model, including its model weights and core evaluation code, to foster industry transparency and collaborative advancement in embodied AI.
- •The Spirit v1.5 model is built on a unified Vision-Language-Action (VLA) architecture that integrates visual perception, language understanding, and action generation into a single end-to-end decision process, which differs from traditional modular pipelines.
- •A key technical focus of Spirit v1.5 is its data collection paradigm, which relies on open-ended, goal-driven diverse data from unscripted real-world experiences rather than highly curated, scripted demonstrations, aiming for more generalizable policies and stable performance on complex robotic tasks.
- •Qianxun Intelligence (also referred to as Qianxun AI) is an embodied AI startup founded in January 2024, focused on developing a 'general brain' for robots, and has recently secured significant funding, including nearly 20 billion yuan (~$2.9 billion) in February 2026 and an additional 1 billion RMB in April 2026.
🛠️ Technical Deep Dive
- Spirit v1.5 VLA Model:
- Developed by Spirit AI, an embodied AI startup.
- Employs a unified Vision-Language-Action (VLA) architecture that integrates visual perception, language understanding, and action generation into a single end-to-end decision process, aiming to reduce information loss and enable more consistent behavior across complex, multi-stage tasks.
- Built upon a Qwen/Qwen3-VL-4B-Instruct Vision-Language Model (VLM) backbone.
- Utilizes a DiT (Diffusion Transformer) action head for generating robot actions.
- Features a policy inference API for execution.
- Trained on open-ended, goal-driven diverse data, capturing a continuous flow of skills from unscripted real-world experiences, which is intended to lead to more generalizable policies and stable performance on complex robotic tasks.
- Evaluated on the RoboChallenge Table30 benchmark, which is a real-hardware evaluation platform supporting 24/7 testing on mainstream robotic systems such as Franka, Arx5, UR5, and ALOHA, assessing capabilities in 3D localization, occlusion handling, temporal reasoning, long-horizon execution, and cross-robot generalization.
- The training script leverages PyTorch's Fully Sharded Data Parallel (FSDP) for efficient multi-GPU training, which shards model parameters across GPUs, reduces memory footprint, and supports gradient checkpointing.
- Open-source release includes model weights and core evaluation code, available on platforms like GitHub and Hugging Face.
- Sunray S600 chip: Insufficient relevant technical data exists for an AI chip named "Sunray S600"; search results primarily refer to UV curing systems.
🔮 Future ImplicationsAI analysis grounded in cited sources
The partnership will accelerate the commercialization of general-purpose embodied AI in robotics.
By integrating a top-performing open-source VLA model with a high-performance, mass-producible chip, the collaboration provides a ready-to-deploy hardware-software solution, lowering barriers for robot manufacturers.
The focus on high-performance edge inference will enable more autonomous and responsive robotic systems in diverse real-world environments.
High-performance edge AI processing allows robots to make real-time decisions locally without constant cloud connectivity, which is crucial for dynamic and unstructured settings.
The open-source nature of Spirit v1.5 will foster a broader ecosystem of developers and applications for the integrated hardware-software solution.
Open-sourcing the VLA model encourages community contributions, independent verification, and wider adoption, potentially leading to rapid innovation and new use cases for the combined platform.
⏳ Timeline
2015
Digua Robotics was established as part of Horizon Robotics.
2024-01
Qianxun AI (Qianxun Intelligence), an embodied AI startup, was founded.
2025-05
Digua Robotics completed a $100 million Series A financing round.
2025-12
Humanoid robot 'Moz,' powered by Spirit v1.5, was deployed at CATL's Zhongzhou manufacturing base.
2026-01
Spirit AI open-sourced its Spirit v1.5 VLA model after it topped the RoboChallenge benchmark.
2026-03
Digua Robotics completed a $120 million Series B1 funding round.
2026-04
Qianxun Intelligence secured a total of 3 billion RMB in funding across two rounds within 30 days.
2026-04
Digua Robotics secured an additional $150 million in Series B2 funding, bringing total to $270 million.
📎 Sources (11)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 36氪 ↗