つながる知能、広がる現実

@inproceedings{Yun2026,

title = {An Open-Source Modular Benchmark for Diffusion-Based Motion Planning in Closed-Loop Autonomous Driving},

author = {Li Yun and Simon Thompson and Yidu Zhang and Ehsan Javanmardi and Manabu Tsukada},

year  = {2026},

date = {2026-09-15},

urldate = {2026-09-15},

booktitle = {The IEEE International Conference on Intelligent Transportation Systems (ITSC2026)

},

address = {Naples, Italy},

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. Project page: },

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{YANG2026,

title = {Collaborative Semantic Occupancy Prediction with Probabilistic Gaussian},

author = {SIBING YANG and Kimihiko Nakano and Manabu Tsukada and Wei WANG and Ehsan Javanmardi},

year  = {2026},

date = {2026-09-15},

booktitle = {The IEEE International Conference on Intelligent Transportation Systems (ITSC2026)},

address = {Naples, Italy},

abstract = {Vehicle-to-vehicle (V2V) collaborative perception enables

multi-agent information sharing and provides a more

comprehensive understanding of the surrounding environment

for semantic occupancy prediction. However, most existing

collaborative semantic occupancy methods rely on dense

grid-based representations, whose communication cost

increases rapidly due to the explicit modeling of large

amounts of empty space.

In this paper, we propose CoPG, a collaborative semantic

occupancy prediction framework built upon a probabilistic

Gaussian representation. The proposed representation

explicitly models non-empty regions while suppressing

redundant primitives in empty space, enabling a compact yet

expressive scene encoding. Moreover, we introduce a

Gaussian communication strategy that selectively transmits

semantically confident and spatially relevant Gaussians,

substantially reducing communication redundancy. For

multi-agent integration, we employ 3D sparse convolution to

implicitly model spatial interactions among neighboring

Gaussians at the Gaussian level.

Experimental results demonstrate that the proposed method

reduces communication volume by 18.5% while improving mIoU

by 11.8% over strong baselines, validating the

effectiveness and communication efficiency of

Gaussian-level fusion for collaborative semantic occupancy

prediction.

The source code will be made publicly available},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Asabe2026,

title = {At the Limits: Sim-to-Real and Community Lessons from the Autonomous Driving AI Challenge},

author = {Yu Asabe and Shintaro Tomie and Taiki Tanaka and Masahiro Kubota and Shinpei Kato and Manabu Tsukada and Hiroshi Esaki},

year  = {2026},

date = {2026-09-15},

booktitle = {The IEEE International Conference on Intelligent Transportation Systems (ITSC2026)},

address = {Naples, Italy},

abstract = {Developing software-defined vehicles (SDV) requires an

integrated mastery of automotive control, software

engineering, and machine learning. As participants in the

Autonomous Driving AI Challenge, we navigated this

multi-domain integration within a high-speed autonomous

racing context. This paper documents our technical journey

in bridging the "sim-to-real" gap, moving from

high-fidelity digital twins to physical EV racing karts

operating at their dynamic limits. A unique feature of this

challenge is its professional-level development

infrastructure, incorporating cloud-native Continuous

Integration/Continuous Delivery (CI/CD) pipelines and

Over-The-Air (OTA) deployment. We detail how this framework

enabled a rapid PDCA (Plan-Do-Check-Act) cycle, allowing us

to transition from virtual validation to real-world

performance tuning in minutes. We detail the methodologies

employed to operate at the vehicle's physical limits,

specifically focusing on the challenges of robust

simulation-to-real transfer and the implementation of

hybrid architectural strategies. Beyond the individual

technical challenges, we highlight how a collaborative

community ecosystem driven by open-source heuristics,

real-time telemetry, and "edutainment" broadcasting

accelerated our technical growth. Our findings serve as a

participant-led case study on the efficacy of autonomous

racing as a catalyst for both technical innovation and the

democratization of SDV engineering expertise. },

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Chauhan2026,

title = {Colored Shared Spaces (CSS): How Visual Design Transforms Pedestrian Experiences},

author = {Vishal Chauhan and Anubhav Anubhav and Chia Ming Chang and Ehsan Javanmardi and Takeo Igarashi and Alex Orsholits and Kantaro Fujiwara and Manabu Tsukada},

year  = {2026},

date = {2026-07-26},

urldate = {2026-07-16},

booktitle = {28th International Conference on Human-Computer Interaction (HCII2026)},

address = {Montreal, Canada},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Aizono2026,

title = {RealSim-CP: A High-Fidelity Multimodal Cooperative Perception Dataset Bridging the Simulator–Real World Gap},

author = {Yuji Aizono and Ehsan Javanmardi and Fardin Ayar and Mahdi Javanmardi and Manabu Tsukada and Hiroshi Esaki},

year  = {2026},

date = {2026-06-09},

booktitle = {2026 IEEE 103rd Vehicular Technology Conference (VTC2026-Spring)},

address = {Nice, France},

abstract = {Cooperative perception, which enables vehicles to share sensory information with other vehicles and roadside infrastructure, is essential for advancing autonomous driving beyond single-vehicle limitations. However, existing cooperative perception datasets suffer from two critical limitations: the lack of representation of Japanese traffic environments with their unique characteristics (e.g., left-hand traffic, distinctive vehicle types, and region-specific infrastructure), and the high cost of real-world data collection that constrains dataset scale and diversity.



To address these challenges, we present RealSim-CP, a novel cooperative perception dataset generated using the Driving Intelligence Validation Platform (DIVP), a physics-based simulation system that employs ray tracing and electromagnetic-wave modeling to produce highly realistic sensor data comparable to real-world quality. Leveraging DIVP’s high-fidelity simulation capabilities, we efficiently generate a large-scale dataset comprising synchronized multimodal data—camera images and LiDAR point clouds—from multiple cooperative agents, including vehicles and roadside units. The dataset covers three urban maps representing Tokyo regions (Aomi, Odaiba, and the Shutoko Expressway) under diverse environmental conditions, including clear daytime, rainy daytime, and clear nighttime scenarios.



All data are provided in the standardized OpenLABEL format with annotations for 12 object classes, totaling 140k images and 30k point clouds. We further evaluate RealSim-CP using CoopDet3D, a state-of-the-art multimodal cooperative 3D object detection framework, demonstrating the effectiveness of the dataset for advanced cooperative perception research. These results indicate that high-fidelity simulation can effectively bridge the gap between simulation and real-world deployment while significantly reducing data collection costs. RealSim-CP provides the first region-specific cooperative perception dataset tailored to Japanese traffic environments.

},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Li2026c,

title = {V2XState: Intent-aware Spatial Basis Attention for Cooperative End-to-end Driving},

author = {Dongyang Li and Ehsan Javanmardi and Manabu Tsukada},

year  = {2026},

date = {2026-06-09},

urldate = {2026-06-09},

booktitle = {2026 IEEE 103rd Vehicular Technology Conference (VTC2026-Spring)},

address = {Nice, France},

abstract = {Cooperative autonomous driving relies on shared perception, yet most planning models implicitly treat spatial relevance as scene-driven and largely invariant to the ego vehicle's intent. We argue that spatial relevance for planning should be intent-conditioned: which regions matter depends on the specific maneuver being executed. This paper proposes V2XState, a framework that operationalizes this insight by using ego states and commands to modulate spatial attention over cooperative features. By incorporating kinematic-aware inductive biases through a state-gated basis attention mechanism, V2XState yields context-sensitive emphasis that aligns planning with current driving intent. We integrate this mechanism into a lightweight planning stack and observe that the resulting attention maps are both interpretable and intent-consistent. Experiments on cooperative driving benchmarks validate that intent-aware spatial attention leads to more maneuver-consistent planning, achieving a 8.2%/7.2% reduction in ADE/FDE relative to state-of-the-art baselines.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}