Generative driving planners can represent diverse ego futures, but a downstream scorer can choose only among sampled candidates; it cannot recover an absent maneuver. We present ForeseeDrive, which turns future scene evolution into a deployable interface for shaping candidate generation. Rather than reconstructing dense future states or scoring a fixed menu, ForeseeDrive derives query-indexed tokens from logged actor motion, occupancy evolution, route interactions, and TTC margins, and predicts the same token space from present context at deployment. The tokens condition trajectory geometry before candidates materialize. A zero-token-gated residual path and candidate-preservation loss limit unnecessary drift, while a TTC-margin guard targets brittle safe-to-unsafe boundary crossings. Under their respective benchmark protocols, ForeseeDrive reaches 89.2 PDMS on NAVSIM navtest, 0.41,m average L2 and 0.05% collision rate on nuScenes validation, and a 76.50 driving score on Bench2Drive. Controlled DiffusionDrive/NAVSIM diagnostics show an exact zero-token no-op, a small aligned-over-shuffled selected-PDM advantage under preservation, mixed oracle-best quality, and localized TTC-boundary failures. Together, the benchmark results and bounded mechanism tests support future-conditioned generation as a practical interface for anticipation. Code is included in the supplementary materials.
Figures & tables
Figure 1: Top: fixed-candidate future scoring acts after a base generator produces candidates, so it cannot create an absent future-aware mode. Bottom: ForeseeDrive predicts compact future tokens from present context and conditions the generator before materialization; a TTC-margin guard rejects brittle near-contact transitions while candidate preservation limits drift.
Figure 2: ForeseeDrive framework. Top: logged future evolution defines teacher tokens for actor motion, occupancy, route interaction, and TTC margin. Middle: a deployable predictor estimates matching tokens from present context and conditions the base generator before sampling. Bottom: candidate preservation and a TTC-margin guard constrain the token-conditioned candidates before standard selection.
Method
Input
NC ↑
DAC ↑
TTC ↑
Comf. ↑
EP ↑
PDMS ↑
VADv2- V8192 (2024)
C & L
97.2
89.1
91.6
100.0
76.0
80.9
Hydra-MDP- V8192 (2024a)
C & L
97.9
91.7
92.9
100.0
77.6
83.0
UniAD (2023)
Camera
97.8
91.9
92.9
100.0
78.8
83.4
LTF (2022)
Camera
97.4
92.8
92.4
100.0
79.0
83.8
PARA-Drive (2024)
Camera
97.9
92.4
93.0
99.8
79.3
84.0
TransFuser (2022)
C & L
97.7
92.8
92.8
100.0
79.2
84.0
Table 1: NAVSIM navtest closed-loop comparison following SeerDrive [ 52 ] Table 1. NC: no at-fault collision; DAC: drivable area compliance; TTC: time-to-collision; Comf.: comfort; EP: ego progress; PDMS: predictive driver model score. C: camera; L: LiDAR. ForeseeDrive is instantiated on a DiffusionDrive backbone; daggered V2-99 Centermask variants in the source table are omitted.
Method
L2 (m) ↓
Collision Rate (%) ↓
1s
2s
3s
Avg.
1s
2s
3s
Avg.
ST-P3 (2022)
1.33
2.11
2.90
2.11
0.23
0.62
1.27
0.71
BEV-Planner (2024b)
0.28
0.42
0.68
0.46
0.04
0.37
1.07
0.49
VAD-Tiny (2023a)
0.46
0.76
1.12
0.78
0.21
0.35
0.58
0.38
LAW (2025a)
0.26
0.57
1.01
0.61
0.14
0.21
0.54
0.30
PARA-Drive (2024)
0.25
0.46
0.74
0.48
0.14
0.23
0.39
0.25
Table 2: nuScenes validation planning comparison following SeerDrive [ 52 ] Table 2. L2 is measured in meters and collision rate in percent; lower is better.
Method
Sensors
Driving Score ↑
Success Rate (%) ↑
Efficiency ↑
Comfort ↑
UniAD-Base (2023)
6 Cameras
45.81
16.36
129.21
43.58
VAD (2023a)
6 Cameras
42.35
15.00
157.94
46.01
Hydra-MDP (2024a)
Front+Rear Cameras
59.95
29.82
186.83
18.62
TCP ∗ (2022)
Front Camera
40.70
15.00
54.26
47.80
TCP-Ctrl ∗ (2022)
Front Camera
30.47
7.27
55.97
51.51
TCP-Traj ∗ (2022)
Front Camera
59.90
30.00
76.54
18.08
Table 3: Closed-loop Bench2Drive comparison under the Bench2Drive evaluation protocol. Sensor configurations follow the original reports.
Variant
Sel. PDM
Δ
Mode chg.
Base/no future
0.176
0.000
0/12
Zero token
0.176
0.000
0/12
Shuffled token
0.369
+0.193
8/12
Aligned token
0.378
+0.202
9/12
Table 4: Frozen-backbone selector probe on a disjoint targeted NAVSIM evaluation subset. Only a tiny future-conditioned mode selector is trained; the DiffusionDrive backbone and candidate geometry are frozen.
Figure 3: Controlled DiffusionDrive/NAVSIM diagnostics. Panels trace the mechanism chain: fixed-candidate reranking has limited deployable gains; zero tokens preserve the base generator; future tokens shift candidates before selection; preservation controls damage; TTC replay localizes failures.
Existing latent world models for autonomous driving have opened a promising path toward future-aware driving intelligence. However, they typically treat future latent states as prediction targets or auxiliary signals, rather than directly conditioning trajectory planning. This can entangle current and future features in latent space. In this work, we propose DriveFuture, a future-aware latent world modeling framework for autonomous driving that explicitly learns planning-oriented foresight by conditioning the current latent state modeling process on future world states. Specifically, during training, the model first predicts future latent world states from the current latent state and ego action, and then refines the prediction against the ground-truth future latent state via cross-attention. The resulting future-aware latent serves as an explicit condition for a diffusion-based trajectory planner. During inference, DriveFuture conditions on the predicted future latent state instead of the ground-truth future state. DriveFuture achieves SOTA performance on the public NAVSIM benchmarks, reaching 55.5 EPDMS on NAVSIM-v2 {\textcolor{blue}{navhard}}, 89.9 EPDMS on NAVSIM-v2 {\textcolor{blue}{navtest}}, and 90.7 PDMS on NAVSIM-v1 {\textcolor{blue}{navtest}}, respectively. These results suggest that the key to latent world modeling lies not merely in simulating future states, but more importantly in conditioning current decision-making on future states. Notably, as of April 2026, DriveFuture ranks 1st on the \href{https://huggingface.co/spaces/AGC2025/e2e-driving-navhard}{NAVSIM-v2 {\textcolor{blue}{navhard}}} leaderboard and achieves SOTA performance on \href{https://huggingface.co/spaces/AGC2024-P/e2e-driving-navtest}{NAVSIM-v1 {\textcolor{blue}{navtest}}}.
Yufeng Hong, Xiaotian Zhou, Yingyan Li +6
Beijing Institute of Technology · Institute of Automation, Chinese Academy of Sciences · Beihang University +5
Existing latent world models are typically optimized for future predictability, yet the resulting representations are not necessarily useful for planning in autonomous driving. Predictions are commonly used for pretraining or auxiliary supervision rather than as direct conditioning signals for trajectory generation. We propose ForeDrive, which learns a planning-relevant latent representation and couples it asymmetrically to a Diffusion Transformer (DiT) planner. The planner consumes multi-horizon latent future representations learned with a JEPA-style world model; planning gradients update the shared online encoder, while stop-gradient routing trains the latent predictor with forecasting losses only. Because predicted futures have varying reliability across horizons and BEV trajectories are misaligned with image tokens, we use gated visual fusion, future-status injection, and Trajectory-Adaptive Bias (TAB) to inject future latents as guidance without overriding the current observation. Trained with pure imitation learning and using only the current front-view image as visual input at inference, ForeDrive attains 89.9 PDMS on NAVSIM v1 and 90.0 one-stage EPDMS on NAVSIM v2, without reinforcement learning or an external trajectory scorer.
Sinuo Wang, Zichong Gu, Yuhan Huang +10
Huazhong University of Science and Technology · Shanghai Zaofu Intelligent Technology Co., Ltd. · Tongji University
Trajectory generation underpins urban applications such as traffic management and urban planning. Within road networks, vehicle movement along road segments is largely deterministic, whereas transitions at intersections are critical--multiple outgoing paths may appear equally plausible, yet they can lead to substantially different downstream routes. Existing generative methods such as diffusion models simulate location sequences and fail short of next transition selection, even producing trajectories that drift off the road network. To fill this gap, we advocate a fundamental shift: moving beyond location simulation toward behavior modeling on the road network. In this paper, we revisit trajectory generation through the lens of generative route planning, and introduce a new graph-constrained generative paradigm, Anticipatory Generative Modeling, which resolves next transition selection by incorporating the Anticipation of how the future route may unfold toward the destination. Concretely, we propose a novel generative model named Foresight that formulates a probabilistic model for an "anticipate--select--update--re-anticipate" loop that enables longer-term foresight while continuing sequential generation along the historical trajectory. Extensive experiments demonstrate that Foresight wins state-of-the-art baselines.
Zhenhao Huang, Li Sun, Philip S. Yu
North China Electric Power University · Beijing University of Posts and Telecommunications · University of Illinois Chicago