Spirit-v1.5 - A basic model of embodied intelligence launched by Qianxun Intelligence

What is Spirit-v1.5?

Spirit-v1.5 is a foundational embodied intelligence model launched by Qianxun Intelligence. The model employs diverse and open data collection methods, abandoning the limitations of traditional "clean data," allowing the model to be exposed to richer actions and scenarios during the pre-training phase, resulting in stronger generalization and adaptability. In the RoboChallenge real-device evaluation, Spirit-v1.5 outperformed previous leading models, becoming the current best embodied intelligence model, providing a new technical path and open-source benchmark for the development of embodied intelligence.

Main features of Spirit-v1.5

• Diverse task executionSpirit-v1.5 can perform a variety of daily tasks, such as flower arranging, putting fruit in a basket, hanging toothbrush cups, applying tape, and cleaning up trash.
• Generalization abilityIt has the ability to quickly adapt to and perform tasks in unseen scenarios by using diverse pre-trained data.
• robustnessIt performs well in complex environments and can cope with challenges such as object occlusion, changes in lighting, and robotic arm collaboration.
• Supports multiple robot platformsThe model is compatible with various robot hardware, such as single-arm robots (Franka, UR5e) and dual-arm robots (ALOHA).
• High-efficiency fine-tuningIt can quickly fine-tune for new tasks, reducing training time and resource consumption.

Technical Principles of Spirit-v1.5

• Diverse data collectionThe model abandons the traditional "clean data" approach, adopting an open and loosely controlled data collection method. Collectors act freely around the task objectives, generating data that includes various atomic skills and real-world temporal relationships.
• Pre-training and fine-tuningThe model is pre-trained on large-scale, diverse data to learn general policies rather than optimal solutions for a single task. When fine-tuned on new tasks, the model can quickly adapt and perform excellently.
• Vision-Language-Action (VLA) Model ArchitectureCombining vision, language, and motion modules enables robots to understand natural language commands, perceive their environment, and perform corresponding actions.
• Reinforcement learning and self-monitoring learningThe model uses reinforcement learning to optimize task success rate, while improving its adaptability to environmental changes through self-supervised learning.

The project address for Spirit-v1.5

• Project official website: https://www.spirit-ai.com/en/blog/spirit-v1-5
• GitHub repository: https://github.com/Spirit-AI-Team/spirit-v1.5
• HuggingFace model libraryhttps://huggingface.co/Spirit-AI-robotics/Spirit-v1.5

Application scenarios of Spirit-v1.5

• Home servicesSpirit-v1.5 can assist with household chores, such as cleaning tabletops, organizing items, and helping with cooking, significantly improving the convenience and efficiency of family life.
• Industrial AutomationThe model is applicable to production line operation, quality inspection, and logistics warehousing, improving the efficiency and accuracy of industrial production by efficiently executing tasks.
• Medical assistanceSpirit-v1.5 optimizes medical processes and provides strong support for healthcare workers in areas such as rehabilitation training, medical supplies management, and surgical assistance.
• Education and ResearchAs a research and teaching tool, Spirit-v1.5 assists in academic research and popularization of education, and promotes the educational development of robotics and artificial intelligence.
• Retail and ServicesSpirit-v1.5 can improve retail operational efficiency and enhance customer experience in store management, customer service, and inventory management.
• Agriculture and HorticultureSpirit-v1.5 can be used for crop management and horticultural care, helping agricultural production and horticultural maintenance and improving operational efficiency.