arXiv · 机器人与计算机视觉· Manav Gagvani, Benjamin Namikas, Sivamurugan Velmurugan, Nysa Kumar, Shrey Sharma, Thomas Peterson, Pratyush Mathur, Hyunseo Chang, Rohan Paleja·· 17 天前AI 评分72
Mechanistic Interpretability for End-to-End Self-Driving 论文发布
Mechanistic Interpretability for End-to-End Self-Driving
AI 导读
arXiv 预印本论文《Mechanistic Interpretability for End-to-End Self-Driving》提出在端到端自动驾驶路径规划模型上训练稀疏自编码器(SAEs),通过无监督聚类发现与路线意图、车速和制动行为对齐的可解释结构,路线意图识别准确率达 98.3%。
来源:arXiv · 机器人与计算机视觉 · arxiv.org