中国初创精神AI的机器人系统 在现实世界的任务中取得了50.33%的成功, 优于美国顶级模型。 Chinese startup Spirit AI's robot system achieved 50.33% success in real-world tasks, outperforming a top U.S. model.

通过其Spirit v1.5系统,中国AI初创公司Spirit AI在RoboChallenge现实世界机器人基准测试中取得了顶尖的成绩,超过了美国领先的机器人模型,在30个涉及物体放置和工具使用等日常任务的物理挑战中,实现了50.33%的任务成功率,这是唯一超过50%的模型. Chinese AI startup Spirit AI has topped the RoboChallenge real-world robotics benchmark, surpassing a leading U.S. model with its Spirit v1.5 system, which achieved a 50.33% task success rate across 30 physical challenges involving everyday tasks like object placement and tool use—the only model to exceed 50%. 该公司设在杭州,拥有开放来源的模型,并使用统一的愿景-语言-行动架构,整合认识、推理和行动。 Based in Hangzhou, the company has open-sourced its model and uses a unified Vision-Language-Action architecture to integrate perception, reasoning, and action. 这一成就标志着在情报化方面迈出了重要一步,尽管广泛部署仍要多年。 The achievement marks a significant step in embodied intelligence, though widespread deployment remains years away. CEO Han Fengtao预测二至三年内会采用范围更广的服务机器人。 CEO Han Fengtao predicts broader service robot adoption within two to three years.