FeaXDrive: Feasibility-aware Trajectory-Centric Diffusion Planning for End-to-End Autonomous Driving
Authors: Baoyun Wang, Zhuoren Li, Ran Yu, Yu Che, Xinrui Zhang, Ming Liu, Jia Hu, Chen Lv, Bo Leng
Organizations: College of Automotive and Energy Engineering, Tongji University, Shanghai, 201804, China · College of Transportation Engineering, Tongji University, Shanghai, 201804, China · School of Mechanical and Aerospace Engineering, Nanyang Technological University, 639798, Singapore
End-to-end diffusion planning has shown strong potential for autonomous driving, but the physical feasibility of generated trajectories remains insufficiently addressed. In particular, generated trajectories may exhibit local geometric irregularities, violate trajectory-level kinematic constraints, or deviate from the drivable area, indicating that the commonly used noise-centric formulation in diffusion planning is not yet well aligned with the trajectory space where feasibility is more naturally characterized. To address this issue, we propose FeaXDrive, a feasibility-aware trajectory-centric diffusion planning method for end-to-end autonomous driving. The core idea is to treat the clean trajectory as the unified object for feasibility-aware modeling throughout the diffusion process. Built on this trajectory-centric formulation, FeaXDrive integrates adaptive curvature-constrained training to improve intrinsic geometric and kinematic feasibility, drivable-area guidance within reverse diffusion sampling to enhance consistency with the drivable area, and feasibility-aware GRPO post-training to further improve planning performance while balancing trajectory-space feasibility. Experiments on the NAVSIM benchmark show that FeaXDrive achieves strong closed-loop planning performance while substantially improving trajectory-space feasibility. These findings highlight the importance of explicitly modeling trajectory-space feasibility in end-to-end diffusion planning and provide a step toward more reliable and physically grounded autonomous driving planners.
Figures & tables
Failure Cause
DiffusionDrive ( Liao et al., 2025 )
ReCogDrive ( Li et al., 2025c )
Drivable-area non-compliance
69.25%
56.59%
At-fault collision
32.42%
45.44%
Both
1.67%
2.03%
Table 1: Failure cause distribution in score-zero planning scenes of reproduced diffusion-based planners. Percentages are computed with respect to the score-zero scenes of each planner.
Figure 1: Overview of the proposed FeaXDrive. Compared with noise-centric diffusion planning, FeaXDrive adopts a trajectory-centric formulation in which the predicted clean trajectory serves as the unified object for feasibility-aware modeling. On this basis, the method combines training-time curvature feasibility regularization and inference-time drivable-area guidance to enhance trajectory-space feasibility throughout the diffusion planning process.
Figure 2: Overall architecture of FeaXDrive. Under a trajectory-centric formulation, the predicted clean trajectory serves as the shared object for feasibility-aware modeling throughout training, inference, and post-training. The method integrates adaptive differentiable curvature-regularized training, drivable-area guidance during reverse diffusion sampling, and feasibility-aware GRPO post-training.
Method type
Model
NC ↑
EP ↑
Comf. ↑
TTC ↑
DAC ↑
PDMS ↑
IL
VADv2 ( Chen et al., 2024a )
97.2
76.0
100
91.6
89.1
80.9
Driving-GPT ( Chen et al., 2025 )
98.9
79.7
95.6
94.9
90.7
82.4
Hydra-MDP ( Li et al., 2024 )
97.9
77.6
100
92.9
91.7
83.0
UniAD ( Hu et al., 2023 )
97.8
78.8
100
92.9
91.9
83.4
PARA-Drive ( Weng et al., 2024 )
97.9
79.3
99.8
93.0
92.4
84.0
TransFuser ( Prakash et al., 2021 )
97.7
79.2
100
92.8
92.8
84.0
Table 2: Main closed-loop planning results on the NAVSIM benchmark. RLFT denotes reinforcement learning fine-tuning; for FeaXDrive, it is instantiated as feasibility-aware GRPO.
Method
DiffusionDrive ( Liao et al., 2025 )
ReCogDrive-IL ( Li et al., 2025c )
ReCogDrive w/GRPO ( Li et al., 2025c )
FeaXDrive-IL (Ours)
FeaXDrive w/FA-GRPO (Ours)
Curvature Viol. Rate ↓
8.59%
8.05%
15.5%
0.88%
2.40%
Table 3: Comparison of curvature violation rates among reproduced diffusion-based planners.
Method
x0-pred
Curvature Regularization
Sampling Guidance
PDMS ↑
EP ↑
DAC ↑
Drivable Area Viol. Rate ↓
Curvature Viol. Rate ↓
Baseline
✗
✗
✗
85.32
79.56
93.84
6.16%
11.36%
Trajectory-centric
✓
✗
✗
86.56
81.26
94.58
5.42%
7.51%
+ Training-time feasibility enhancement
✓
✓
✗
86.57
81.32
94.94
5.06%
0.13%
FeaXDrive-IL
✓
✓
✓
88.75
83.34
97.46
2.54%
0.88%
Table 4: IL-stage ablation study of the proposed method on benchmark performance and trajectory-space feasibility.
Figure 3: Comparison of curvature violation counts under different IL-stage ablation settings.
Figure 4: Comparison of drivable-area violation counts under different IL-stage ablation settings.
Method
PDMS ↑
EP ↑
DAC ↑
Drivable Area Viol. Rate ↓
Curvature Viol. Rate ↓
FeaXDrive-IL
88.75
83.34
97.46
2.54%
0.88%
FeaXDrive w/GRPO
90.56
85.13
98.28
1.72%
5.79%
FeaXDrive w/FA-GRPO
90.00
84.20
98.31
1.69%
2.40%
Table 5: Comparison of post-training fine-tuning strategies for FeaXDrive-IL. FA-GRPO denotes feasibility-aware GRPO.
Figure 5: Latency breakdown of FeaXDrive inference.
Figure 6: Qualitative comparison between the noise-centric baseline and FeaXDrive on representative planning scenes. From top to bottom: trajectory-level kinematic infeasibility, local geometric irregularities, and drivable-area inconsistency. Compared with the baseline, FeaXDrive produces trajectories that are smoother, more kinematically feasible, and better aligned with the drivable area.
Diffusion-based planning generates diverse action sequences or trajectories by progressively denoising candidates conditioned on goals, scenes, and constraints.
End-to-End (E2E) autonomous driving models have shown growing capability in recent years, with performance improving on increasingly challenging benchmarks. However, modern generative E2E planners still suffer from a substantial number of catastrophic failures in safety-critical scenarios. We find that many such failures arise from violations of physical constraints and safety requirements, leading to unsafe behavior. Motivated by this finding, in this paper, we focus on improving safety outcomes in generative end-to-end driving with a targeted reduction of catastrophic planning failures, instead of enhancing average planning quality. Towards this end, we propose DriveSafer, a failure-aware safety framework for end-to-end planners. DriveSafer explicitly steers generative planners towards safe behaviors leveraging both training-time safety constraints and inference-time safety guidance. Compared to the state-of-the-art DiffusionDrive model, on the NAVSIM benchmark, DriveSafer reduces the number of catastrophic failures (PDMS=0) by 48%, with over 65% reduction in drivable-area compliance failures.
Diffusion models have become a popular choice for decision-making tasks in robotics, and more recently, are also being considered for solving autonomous driving tasks. However, their applications and evaluations in autonomous driving remain limited to simulation-based or laboratory settings. The full strength of diffusion models for large-scale, complex real-world settings, such as End-to-End Autonomous Driving (E2E AD), remains underexplored. In this study, we conducted a systematic and large-scale investigation to unleash the potential of the diffusion models as planners for E2E AD, based on a tremendous amount of real-vehicle data and road testing. Through comprehensive and carefully controlled studies, we identify key insights into the diffusion loss space, trajectory representation, and data scaling that significantly impact E2E planning performance. Moreover, we also provide an effective reinforcement learning post-training strategy to further enhance the safety and robustness of the learned planner. The resulting diffusion-based learning framework, Hyper Diffusion Planner (HDP), is deployed on a real-vehicle platform and evaluated across 6 urban driving scenarios and 200 km of real-world testing, achieving a notable 10x performance improvement over the base model. Our work demonstrates that diffusion models, when properly designed and trained, can serve as effective and scalable E2E AD planners for complex, real-world autonomous driving tasks.
Yinan Zheng, Tianyi Tan, Bin Huang +11
Institute for AI Industry Research (AIR), Tsinghua University · Xiaomi EV
Autonomous driving in complex traffic requires planners that generalize beyond hand-crafted rules, motivating data-driven approaches that learn behavior from expert demonstrations. Diffusion-based trajectory planners have recently shown strong closed-loop performance by iteratively denoising a full-horizon plan, but they remain difficult to certify and can fail catastrophically in rare or out-of-distribution scenarios. To address this challenge, we present PC-Diffuser, a safety augmentation framework that embeds a certifiable, path-consistent barrier-function structure directly into the denoising loop of diffusion planning. The key idea is to make safety an intrinsic part of trajectory generation rather than a post-hoc fix: we enforce forward invariance along the rollout while preserving the diffusion model's intended path geometry. Specifically, PC-Diffuser (i) evaluates collision risk using a capsule-distance barrier function that better reflects vehicle geometry and reduces unnecessary conservativeness, (ii) converts denoised waypoints into dynamically feasible motion under a kinematic bicycle model, and (iii) applies a path-consistent safety filter that eliminates residual constraint violations without geometric distortion, so the corrected plan remains close to the learned distribution. By injecting these safety-consistent corrections at every denoising step and feeding the refined trajectory back into the diffusion process, PC-Diffuser enables iterative, context-aware safeguarding instead of post-hoc repair...
Eugene Ku, Yiwei Lyu
Department of Computer Science and Engineering, Texas A&M University, College Station, TX, USA