ACE Robotics 发布 ACE-CaP-0:AI 智能体自动编写可复用机器人代码
ACE Robotics' ACE-CaP-0 Has AI Agents Write Reusable Robot Code, Reporting 92.1% Success
ACE Robotics 把机器人技能变成智能体写的代码,240 次真机测试成了 221 次。跟端到端 VLA 模型路线不一样,代码能复用、能审读,做机器人落地的工程师可以看看这套流程怎么搬到自己项目里。
ACE Robotics 推出 ACE-CaP-0,核心思路是 Code as Policies:由 AI 智能体为机器人编写控制代码,随后将代码冻结为可复用的技能模块。作者团队的真机测试显示,240 次真实机器人试验中有 221 次成功,成功率 92.1%。
其工作流程分为两步:智能体先根据任务描述生成控制代码,代码通过验证后即被固化为技能,后续相似任务直接调用而无需重新生成。这种方式将大模型的推理开销集中在首次编写阶段,执行阶段依赖确定性的代码,减少了每次动作都调用模型带来的延迟与不稳定性。
92.1% 的数字来自作者自报的测试,尚未经过第三方验证,但这套“代码即技能”的路线与端到端 VLA 模型形成互补:前者强调可复用与可解释,后者强调泛化。对做机器人部署的工程团队来说,技能库沉淀的做法可能更容易落地到重复性任务场景。
ACE Robotics' ACE-CaP-0 Has AI Agents Write Reusable Robot Code, Reporting 92.1% Success
ACE Robotics' ACE-CaP-0 has an AI agent write and freeze robot control code. The authors' own tests report 221 of 240 successful real-robot trials, or 92.1%.
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include ACE-CaP-0, Code as Policies, agents, embodied AI.
Primary source verification: published directly by pandaily with first-party authority.