RoboSPA Benchmark for VLA Models
RoboSPA provides a comprehensive evaluation framework for robotic AI systems, addressing critical gaps in current benchmarks for complex spatial reasoning and long-horizon planning.
RoboSPA provides a comprehensive evaluation framework for robotic AI systems, addressing critical gaps in current benchmarks for complex spatial reasoning and long-horizon planning.
这篇用一个可控环境拆解了AI长程规划的训练机制,发现次优数据会拖累性能,多教师蒸馏能整合不同能力,值得做训练方法的人细读。
想看看你用的LLM在多工具长流程场景下到底多靠谱?PlanBench-XL用上千个工具设计了真实任务链,测出来主流模型成功率不到40%,值得一测。