Research KB 登录

EVI-031 Evidence

Table 1/2 + 正文:G2DP† reactive +3.7、interPlan +0.81、精修依赖 +4.1 vs DP +12.8。

id
updated
type evidence
claims CLM-031, CLM-032
source PPR-2606.26017
observation Table 1 (Test14-hard R: G2DP 73.91 vs G2DP† 77.61) + Table 2 (interPlan overall 61.74 vs 62.55) + §Results: refinement dependency +4.1 vs Diffusion Planner +12.8
method G2DP† removes neighbor-agent history stacks from model input (current states only); same training pipeline; compare NR/R splits and +ref. deltas
scope reactive/interactive settings; NR splits favor history-carrying G2DP on several metrics

Result

Test14-hard R:G2DP† 77.61 vs G2DP 73.91(+3.70);Test14 R:88.42 vs 85.26;Val14 R:86.67 vs 83.91。interPlan overall:62.55 vs 61.74。精修依赖:Test14-hard NR 收敛后 G2DP\({}^\dagger\)+ref. 相对裸模型 +4.1,Diffusion Planner 相对裸模型 +12.8。NR 侧反向:Val14 NR G2DP 92.26 > G2DP† 91.08。

Caveats

“消除时间相关性”的归因为作者解释(历史特征在反应式重演下分布漂移),无直接特征层证据;interPlan 差距(+0.81)小于 Test14-hard R(+3.70),增益幅度因基准而异。

关联(12)

  • PPR-2501.15564 Diffusion-Based Planning for Autonomous Driving with Flexible Guidance
  • PPR-2606.26017 G2DP: Diffusion Planning with Spatio-Temporal Grid Guidance
  • FRM-2606.26017 G2DP 五层重建:核心对象是 CPT-013 时空代价体(占据概率栅格 × 路线进度图融合)在推理期的可微引导;公理层为能量重加权目标分布、Top-K 车体聚合与末段注入等六个作者前提;论证主干是 Table 4 在同一 DiT 骨干下隔离『稠密网格 vs 稀疏实例』的引导模态消融(稀疏臂按 PPR-2501.15564 附录 C 实现、以 oracle 邻车供给)。
  • CLM-031 去掉邻车历史输入的 history-free 变体在 reactive/interactive 闭环下反而更优,且加事后规则精修后的增益依赖远小于稀疏引导对照(作者推断);该优势严格限于 reactive/interactive 设定,NR 下不成立。
  • CLM-032 稠密网格引导是纯推理期模块:占据预测器与生成骨干分开训练、仅在推理耦合,不重训不改骨干即可注入解析梯度,且部分替代事后规则精修(作者推断)。
  • EVI-024 co-source Table 1:G2DP 系居纯 IL planner 首位,Test14-hard R 77.61(+7.2 over Flow Planner,+8.4 over Diffusion Planner)。
  • EVI-025 co-source Table 4:Grid +3.35 score/+3.59 collision/+4.94 TTC;Sparse(Oracle) −1.96 score/−10.13 comfort/+0.64 collision。稀疏臂按 PPR-2501.15564 附录 C 式 10 的实例级有向距离能量实现,与稠密臂在信号形态与信息内容两轴同时不同(评注已补)。
  • EVI-026 co-source Tables 2–3:interPlan 零样本 G2DP† 62.55(+9.65),collision +10.15、TTC +10.75;progress/speed limit 略降。
  • EVI-027 co-source Table 5:G2DP 较 DP collision +11.0、progress +10.5、TTC +6.7;progress/along-route 超 PDM-Closed。
  • EVI-028 co-source Fig. 8:steps 8–9 + lambda_t=0.5 峰值 77.61;早窗与 lambda_t≥0.8 均退化。
  • EVI-029 co-source Table 6:Top-15 双split峰值 92.26/80.05;Average 86.33 洗信号、Max 76.84 抗伪影差。
  • EVI-030 co-source Fig. 7:γ=0.95 峰值;γ=1.0 过度保守刹车致高密度区追尾,γ=0.2 安全弱化。