An Open-Source Modular Benchmark for Diffusion-Based Motion Planning in Closed-Loop Autonomous Driving

Yun Li; Simon Thompson; Yidu Zhang; Ehsan Javanmardi; Manabu Tsukada

cs.ROMar 1, 2026

An Open-Source Modular Benchmark for Diffusion-Based Motion Planning in Closed-Loop Autonomous Driving

Authors: Yun Li, Simon Thompson, Yidu Zhang, Ehsan Javanmardi, Manabu Tsukada

Organizations: Graduate School of Information Science and Technology, The University of Tokyo, Tokyo, Japan. · TIER IV, Inc., Tokyo, Japan.

Abstract

Diffusion-based motion planners have achieved state-of-the-art results on benchmarks such as nuPlan, yet their evaluation within closed-loop production autonomous driving stacks remains largely unexplored. Existing evaluations abstract away ROS 2 communication latency and real-time scheduling constraints, while monolithic ONNX deployment freezes all solver parameters at export time. We present an open-source modular benchmark that addresses both gaps: using ONNX GraphSurgeon, we decompose a monolithic 18,398 node diffusion planner into three independently executable modules and reimplement the DPM-Solver++ denoising loop in native C++. Integrated as a ROS 2 node within Autoware, the open-source AD stack deployed on real vehicles worldwide, the system enables runtime-configurable solver parameters without model recompilation and per-step observability of the denoising process, breaking the black box of monolithic deployment. Unlike evaluations in standalone simulators such as CARLA, our benchmark operates within a production-grade stack and is validated through AWSIM closed-loop simulation. Through systematic comparison of DPM-Solver++ (first- and second-order) and DDIM across six step-count configurations (N in {3, 5, 7, 10, 15, 20}), we show that encoder caching yields a 3.2x latency reduction, and that second-order solving reduces FDE by 41% at N=3 compared to first-order. The complete codebase will be released as open-source, providing a direct path from simulation benchmarks to real-vehicle deployment.

Figures & tables

Fig. 1: (a) Monolithic ONNX graph (18,398 nodes) compiled to TensorRT. (b) Modular decomposition: three independently executable modules with the DPM-Solver++ loop in C++; the context encoder runs once and is cached across all denoising steps.

† Encoder (3,417) + DiT core (1,237) + Turn indicator (7).
	Monolithic	Modular (ours)
ONNX nodes	18,398	4,661 †
Model size	47 MB	46 MB
Enc. calls / cycle	$N{+}1$	1 (cached)
Runtime backend	TensorRT (GPU)	ONNX Runtime (CPU/GPU)
Solver config	Frozen at export	Runtime-configurable

TABLE I: Monolithic vs. modular architecture.

Fig. 2: Closed-loop evaluation environment. (a) RViz visualization showing the HD map, point cloud, and planned trajectory within the Autoware stack. (b) AWSIM 3D simulation at an urban intersection in the Nishishinjuku map, with the ego vehicle and sensor perception rays.

Fig. 3: Denoising progression of the 8s-horizon waypoint. (a) Per-step predicted position in normalized coordinates: longitudinal ( blue ) and lateral ( orange ); three phases are visible: coarse (steps 1–2), refinement (3–6), and fine adjustment (7–10). (b) Relative error to the converged step-10 prediction.

Fig. 4: Latency–accuracy Pareto frontier. Horizontal gap: encoder caching benefit; vertical gap: solver order benefit. Dashed line: 100 ms planning budget.

Fig. 5: Solver comparison. (a) FDE and (b) ADE vs.

N{=}10

p{=}2

reference. Shaded:

\pm 1\sigma

† $N{+}1$ : $N$ solver steps + 1 final denoise-to-zero (Alg. 1, line 10).
Component	Time	Mono. calls	Mod. calls
Context Encoder	27.5 ms	$N{+}1$	1
DiT Core † (per step)	2.3 ms	$N{+}1$	$N{+}1$
C++ Solver overhead	0.01 ms/step	—	$N$
Total ( $N{=}10$ )		328 ms	53 ms

TABLE II: Inference latency breakdown.

Explore similar work

One Diffusion Model, Two Roles: Guided Trajectory Planning and Safety-Critical Scenario Generation in Closed-Loop Simulation
Sep 4, 2026Arka Pal, Rajesh Kumar, Hannes Eriksson +4Diffusion-Based Planning Diffusion Model Guidance
KTH Royal Institute of Technology, Sweden · Zenseact AB, Sweden · Chalmers University of Technology, Sweden
MISTY: High-Throughput Motion Planning via Mixer-based Single-step Drifting
Apr 23, 2026Yining Xing, Zehong Ke, Yiqian Tu +3Trajectory Generation Diffusion-Based Planning
School of Vehicle and Mobility, Tsinghua University, Beijing, China · State Key Laboratory of Intelligent Green Vehicle and Mobility, Tsinghua University, Beijing, China
Muninn: Your Trajectory Diffusion Model But Faster
May 11, 2026Gokul Puthumanaillam, Hao Jiang, Ruben Hernandez +4Trajectory Generation Diffusion-Based Planning
University of Illinois Urbana-Champaign. · Florida International University. · Providence College.

Sep 4, 2026J/K move · Enter open · S save

One Diffusion Model, Two Roles: Guided Trajectory Planning and Safety-Critical Scenario Generation in Closed-Loop Simulation

Arka Pal, Rajesh Kumar, Hannes Eriksson, Rémi Lacombe+3

KTH Royal Institute of Technology, Sweden · Zenseact AB, Sweden · Chalmers University of Technology, Sweden

Diffusion-Based Planning Diffusion Model Guidance Autonomous Driving Scenario Generation Diffusion Transformer Autonomous Driving Safety Evaluation Adversarial Scenario Generation

Open paper AskPDF arXiv

Diffusion probabilistic models can capture the multi-modal, interaction-rich distribution of joint future trajectories in driving scenes. We show that a single pretrained diffusion traffic model can serve two complementary roles in the autonomous driving development loop: as an ego motion planner, and as a controllable generator of safety-critical scenarios for stress-testing the planners. On the planning side, we introduce a Single-Stream Dual-Stream (SSDS) diffusion-transformer decoder that fuses scene context via joint attention rather than late cross-attention, improving closed-loop performance on nuPlan. We further propose Decoupled Annealing Posterior Sampling with Energy (DAPSE), a training-free guidance scheme that injects arbitrary energy functions at the clean-sample level, avoiding the first-order approximation errors while requiring no auxiliary networks. Beyond planning, we leverage the same diffusion model as a controllable scenario generator to create realistic long-tail driving interactions for closed-loop evaluation. Through inference-time guidance, selected agents are steered toward safety-critical behaviors, including aggressive cut-ins, lead-vehicle braking, and combined longitudinal-lateral interactions, while preserving realistic traffic behaviors. Evaluated in closed-loop nuPlan simulations with independent black-box planners, the generated scenarios expose failure modes that remain hidden under standard benchmarks. Although the SSDS-based planner achieves stronger nominal performance, it experiences larger degradation under these challenging scenarios, demonstrating that benchmark superiority does not necessarily translate to robustness. These results demonstrate that a single learned traffic prior can simultaneously improve motion planning and provide a realistic framework for systematic planner robustness evaluation.

Abstract

Explore similar work

One Diffusion Model, Two Roles: Guided Trajectory Planning and Safety-Critical Scenario Generation in Closed-Loop Simulation

MISTY: High-Throughput Motion Planning via Mixer-based Single-step Drifting

Muninn: Your Trajectory Diffusion Model But Faster