End-to-end multi-modal planning has been widely adopted to model the uncertainty of driving behavior, typically by scoring candidate trajectories and selecting the optimal one. Existing approaches generally fall into two categories: scoring a large static trajectory vocabulary, or scoring a small set of dynamically generated proposals. While static vocabularies often suffer from coarse discretization of the action space, dynamic proposals provide finer-grained precision and have shown stronger empirical performance on existing benchmarks. However, it remains unclear whether dynamic generation is fundamentally necessary, or whether static vocabularies can already achieve comparable performance when they are sufficiently dense to cover the action space. In this work, we start with a systematic scaling study of Hydra-MDP, a representative scoring-based method, revealing that performance consistently improves as trajectory anchors become denser, without exhibiting saturation before computational constraints are reached. Motivated by this observation, we propose SparseDriveV2 to push the performance boundary of scoring-based planning through two complementary innovations: (1) a scalable vocabulary representation with a factorized structure that decomposes trajectories into geometric paths and velocity profiles, enabling combinatorial coverage of the action space, and (2) a scalable scoring strategy with coarse factorized scoring over paths and velocity profiles followed by fine-grained scoring on a small set of composed trajectories. By combining these two techniques, SparseDriveV2 achieves 92.0 PDMS and 90.1 EPDMS on NAVSIM, with 89.15 Driving Score and 70.00 Success Rate on Bench2Drive with a lightweight ResNet-34 as backbone. Code and model are released at https://github.com/swc-17/SparseDriveV2.
Figures & tables
#Anchors
EPDMS
Memory (MB)
1024
85.02
9531
2048
85.80 (+0.78)
11451
4096
86.33 (+0.53)
15513
8192
86.78 (+0.45)
23261
16384
87.35 (+0.57)
38877
32768
–
OOM
Table 1: Systematic scaling study of static vocabulary density based on Hydra-MDP [ 18 ] . We report EPDMS on NAVSIM v2 and training memory consumption on NVIDIA L20 GPU with 48 GB memory. EPDMS improvements are computed relative to the previous scale.
Figure 1: Overview of the SparseDriveV2 framework. SparseDriveV2 factorizes (a) a spatiotemporal trajectory into (b) a geometric path and a velocity profile, and reconstructs the trajectory by (c) composing the two components. This representation enables (d) a super-dense trajectory vocabulary constructed from a compact set of paths and velocity profiles. Conditioned on (e) scene features, the scalable scoring strategy first performs (f) coarse factorized scoring over paths and velocity profiles to select top- k candidates, followed by (g) fine-grained scoring over the composed trajectories to produce the final planning decision.
Method
Img. Backbone
NC ↑
DAC ↑
TTC ↑
Comf. ↑
EP ↑
PDMS ↑
VADv2 [ 1 ]
ResNet-34
97.2
89.1
91.6
100
76.0
80.9
UniAD [ 8 ]
ResNet-34
97.8
91.9
92.9
100
78.8
83.4
Transfuser [ 2 ]
ResNet-34
97.7
92.8
92.8
100
79.2
84.0
PARA-Drive [ 31 ]
ResNet-34
97.9
92.4
93.0
99.8
79.3
84.0
DRAMA [ 35 ]
ResNet-34
98.0
93.1
94.8
100
80.1
85.5
GoalFlow [ 33 ]
ResNet-34
98.3
93.8
94.3
100
79.8
85.7
Table 2: Performance on NAVSIM v1 navtest leaderboard.
Method
Img. Backbone
NC ↑
DAC ↑
DDC ↑
TL ↑
EP ↑
TTC ↑
LK ↑
HC ↑
EC ↑
EPDMS ∗ ↑
EPDMS ↑
Ego Status MLP
ResNet-34
93.1
77.9
92.7
99.6
86.0
91.5
89.4
98.3
85.4
64.0
–
Transfuser [ 2 ]
ResNet-34
96.9
89.9
97.8
99.7
87.1
95.4
92.7
98.3
87.2
76.7
–
Hydra-MDP++ [ 16 ]
ResNet-34
97.2
97.5
99.4
99.6
83.1
96.5
94.4
98.2
70.9
81.4
–
DriveSuprim [ 34 ]
ResNet-34
97.5
96.5
99.4
99.6
88.4
96.6
95.5
98.3
77.0
83.1
–
ARTEMIS [ 5 ]
ResNet-34
98.3
95.1
98.6
99.8
81.5
97.4
96.5
98.3
98.3
83.1
–
DiffusionDriveV2
ResNet-34
97.7
96.6
99.2
99.8
88.9
97.2
96.0
97.8
91.0
85.5
87.5
Table 3: Performance on NAVSIM v2 navtest leaderboard. EPDMS ∗ denotes results computed using the original NAVSIM v2 evaluation code, in which human-behavior filtering was not applied when calculating EPDMS. This version is reported for fair comparison with prior works. EPDMS denotes results computed using the updated official implementation after the issue was resolved.
Table 5: Multi-Ability results on Bench2Drive.* denotes expert feature distillation.
Scalable Vocabulary
Scalable Scoring
Np
Nv
EPDMS ↑
Path Attn
Re-cond.
EPDMS ↑
512
128
88.7
MHA
✗
87.7
512
256
89.2
MHA
✓
89.9
1024
128
89.5
DFA
✗
89.9
1024
256
90.1
DFA
✓
90.1
Table 6: Ablation studies on NAVSIM v2. Left: Effect of scaling the factorized vocabulary size by varying the number of path anchors ( Np ) and velocity anchors ( Nv ). Right: Effect of the scalable scoring design, including path–scene interaction type and trajectory re-conditioning. MHA denotes multi-head cross-attention and DFA denotes deformable aggregation.
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 2: SparseDriveV2 produces smoother trajectory than the baseline method in sharp-turning scenarios.
Figure 3: SparseDriveV2 achieves higher traffic efficiency, while the baseline method remains stationary.
Figure 4: SparseDriveV2 is better aligned with the expert trajectory in terms of high-level intent, enabled by geometric path modeling.
Figure 5: Failure cases: SparseDriveV2 generates trajectories with incorrect navigation decisions in certain scenarios, possibly due to insufficient navigation information.
Multi-modal planning is promising for autonomous driving by representing multiple plausible behaviors in ambiguous and long-tail scenarios. Existing methods mainly focus on improving trajectory multi-modality, enhancing trajectory representations, or reshaping the candidate distribution. Nevertheless, we identify a pronounced generation-evaluation asymmetry in multi-modal planning: despite strong oracle performance, existing planners often fail to reliably select the best available candidate, leaving substantial planning potential unrealized. To address this challenge, we propose iDriveVLA, a multi-modal planning framework that improves the candidate trajectory space while enabling more reliable and context-aware trajectory evaluation. Specifically, iDriveVLA introduces a unified trajectory evaluator comprising a Safety-aware Scorer for quality and risk estimation, together with a VLM-guided Modulator for scene-adaptive criterion weighting. We further develop an oracle-aligned progressive training strategy consisting of candidate imitation pretraining, candidate space refinement, and semantic ranking alignment. On the public NAVSIM v1 leaderboard, iDriveVLA achieves a new state-of-the-art performance of 94.95 PDMS, surpassing the human-expert reference.
Zeyu He, Shiqi Liu, Ke Chen +14
School of Vehicle and Mobility & College of AI, Tsinghua University · Dongfeng Motor Corporation Research & Development Institute
While multimodal large language models (MLLMs) provide advanced reasoning for autonomous driving, translating their discrete semantic knowledge into continuous trajectories remains a fundamental challenge. Existing methods often rely on unimodal planning heads that inherently limit their ability to represent multimodal driving behavior. Furthermore, most generative approaches frequently condition on one-hot encoded actions, discarding the nuanced navigational uncertainty critical for complex scenarios. To resolve these limitations, we introduce LAD-Drive, a generative framework that structurally disentangles high-level intention from low-level spatial planning. LAD-Drive employs an action decoder to infer a probabilistic meta-action distribution, establishing an explicit belief state that preserves the nuanced intent typically lost by one-hot encodings. This distribution, fused with the vehicle's kinematic state, conditions an action-aware diffusion decoder that utilizes a truncated denoising process to refine learned motion anchors into safe, kinematically feasible trajectories. Extensive evaluations on the LangAuto benchmark demonstrate that LAD-Drive achieves state-of-the-art results, outperforming competitive baselines by up to 59% in Driving Score while significantly reducing route deviations and collisions. We will publicly release the code and models on https://github.com/iis-esslingen/lad-drive.
Fabian Schmidt, Karol Fedurko, Markus Enzweiler +1
Institute for Intelligent Systems, Esslingen University of Applied Sciences, Germany. · Department of Computer Science, University of Freiburg, Germany.
Multimodal driving planning faces a long-standing tension between two paradigms: scoring-based methods benefit from dense reward supervision but are confined to a fixed action vocabulary, while anchor-based methods generate proposals dynamically yet suffer from sparse supervision constrained to a single ground-truth trajectory. In this work, we propose FlowR2A, which resolves this tension by reframing simulation-based rewards from discriminative targets into generative conditions. By learning the reward-conditioned action distribution from dense trajectory-reward pairs with a flow-matching decoder, FlowR2A unifies the dense supervision of scoring-based methods with the proposal generation of anchor-based methods in a single generative model, forcing the model to internalize the correlation between an action and its outcomes in safety, progress, comfort, and rule compliance. To balance hard safety constraints against soft progress objectives, we introduce fine-grained per-timestep reward conditioning and reward noise augmentation. The generative formulation naturally supports controllable test-time sampling via reward guidance and anchored sampling, producing high-quality proposals. FlowR2A achieves state-of-the-art results on the NAVSIM v1 and v2 benchmarks, with multimodal proposals of substantially higher quality than prior methods.
Xirui Li, Zhe Liu, Xiaoqing Ye +4
1The University of Hong Kong · 2Changan Automobile