---
title: "G-MARK：基于知识图谱的多智能体协同驾驶推理框架"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.19964v1"
permalink: "https://feed.astralor.com/items/itm_e7998e6f596bee133396fd4c"
published: "2026-08-20T12:35:12.000Z"
score: 77
channel: "ai"
tags: ["G-MARK","V2V 协同","多智能体推理","知识图谱","自动驾驶"]
---

# G-MARK：基于知识图谱的多智能体协同驾驶推理框架

## AI 点评

> 这篇论文值得关注的点在于，它把多智能体协同从“压缩特征”拉回到“显式知识”，让系统能回答谁看到了什么，这类可追溯性对安全攸关的自动驾驶非常关键。对做车联网协同感知或可解释 AI 的研究者，尤其是关心遮挡场景和通信效率的人，很有参考价值。

## 速览

G-MARK 将多车协同感知建模为带来源归属的知识图谱，显式保留每个目标由哪辆车观测、自车可见性、不确定性、冲突证据、空间关系与规划上下文，并用轻量任务头完成目标推理、运动预测、控制选择和轨迹预测。相比基线，遮挡推理准确率提升42.2%，控制选择误差降低13.1%，结构化通信负载缩小25.6倍。

## 关键点

- 提出显式可解释的协同驾驶推理框架，将多车观测转换为带来源的知识图谱。
- 保留遮挡、冲突与不确定性等关键信息，支撑物体推理、运动预测、控制选择和轨迹预测。
- 相比 SOTA 基线，遮挡推理准确率提升 42.2%，控制选择误差降低 13.1%。
- 结构化通信负载缩小 25.6 倍，同时保持轨迹预测精度相当。

## 全文

Autonomous driving systems must operate under partial observability, where safety-critical objects may be occluded or visible only to neighboring connected vehicles. Vehicle-to-vehicle cooperation can reduce this uncertainty, but existing cooperative driving methods often compress multi-agent evidence into latent features or hidden multimodal states. As a result, they obscure which agent observed each object, whether the object is visible to the ego vehicle, and how conflicting evidence affects downstream decisions. We propose G-MARK, a grounded multi-agent reasoning framework that converts cooperative object-centric observations into explicit provenance-aware knowledge graphs (KGs). The resulting KGs preserve object hypotheses together with their source attribution, ego-versus-partner visibility, uncertainty, conflicts, spatial relations, and planning-relevant context. G-MARK then derives a shared feature representation from these KGs, enabling lightweight task heads to support object reasoning, motion prediction, control selection, and trajectory forecasting. Compared with the state-of-the-art baseline, GMARK improves occlusion reasoning accuracy by 42.2%, reduces control-selection error by 13.1%, and achieves comparable trajectory-planning accuracy with a 25.6x smaller structured communication payload. Our code is available at https://github.com/bhavyagupta98/g-mark.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_e7998e6f596bee133396fd4c