つながる知能、ひろがる現実

@inproceedings{Gong2026b,

title = {CARE: Camera-Residual Reserves for First Sightings in Adaptive LiDAR Sensing},

author = {Jiachen Gong and Yun Li and Ehsan Javanmardi and Wencan Mao and Manabu Tsukada},

year  = {2026},

date = {2026-12-14},

urldate = {2026-12-14},

booktitle = {The 18th Asian Conference on Computer Vision (ACCV 2026)},

address = {Osaka, Japan},

abstract = {Adaptive LiDAR scanning, which concentrates a limited sensing budget on informative regions of interest based on past object tracks, is becoming a key enabler in autonomous driving by lowering data volume while maintaining detection accuracy. However, existing scanning policies face three key challenges. First, history-driven approaches depend heavily on past object tracks, leading to delayed or missed detection of unseen objects. Second, methods leveraging random or uniform sampling outside the predicted regions have no awareness of where new objects appear. Third, camera-guided alternatives spend their budget on all camera detections, resampling objects that are already covered, which costs overall recall in crowded scenes and detection range when budgets are scarce. To address these limitations, this paper introduces the CAmera-REsidual reserve (CARE), a training-free allocation rule that reserves part of a fixed ray budget for the directions of current camera detections that the track forecasts cannot explain, while the remaining budget follows the base history policy and unused reserve returns to a shared random floor. The paper makes three key contributions: First, a leakage-free ray-budget evaluation on nuScenes (150 scenes, 4,148 events) that measures the first-sighting loss of history-driven scanning, with a strict-causal variant using the preceding keyframe. Second, CARE raises first-sighting recall by 5.2, 5.2, and 4.3 points at 10%, 20%, and 35% budgets over the history policy with paired intervals excluding zero; the camera cue drives this gain, and the first-sighting versus overall trade-off is reported openly as a budget-dependent Pareto choice. Third, a safety-bounded forgetting module that releases budget from receding or static tracks beyond a speed-dependent guard distance; at tight budgets forgetting without the guard significantly harms near-field recall, and the guard is what keeps forgetting safe at the tightest budget. The pipeline runs end to end on a real vehicle and, in a closed-loop simulation, detects an occluded pedestrian earlier and brakes more reliably than history-driven scanning.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Chi2026b,

title = {Blockage-Aware Non-stationary Dynamic Bandit for User Association in mmWave V2X Networks},

author = {Weiqi Chi and Manabu Tsukada},

year  = {2026},

date = {2026-12-07},

urldate = {2026-12-07},

booktitle = {IEEE Global Communications Conference (GLOBECOM 2026)},

address = {Macau, China},

abstract = {In millimeter-wave (mmWave) vehicular networks, dense base station (BS) deployments and dynamic blockages introduce severe non-stationarity into the user association (UA), posing a critical mobility management challenge that hinders rapid and effective UA decisions. Traditional Multi-Armed Bandit (MAB) frameworks assume stationary reward distributions and fail to handle the rapid context-reward mapping shifts caused by vehicle mobility and transient blockages. To address this, we propose Blockage-Aware Non-stationary Dynamic Bandit (BAND), a fully distributed, Channel state information (CSI)-free mobility management framework for mmWave vehicular networks, formulating UA as a non-stationary contextual bandit problem and enables online adaptive optimization without requiring central coordination or offline training. BAND employs a cumulative sum-based change detection (CUSUM-CD) to dynamically narrow the active BS set, reducing exploration overhead while tracking reward distribution shifts. Proactive blockage detections suppress transient signal degradation in reward estimation. Simulations demonstrate over 44% regret reduction and up to 27.3% network communication rate improvement over hypercube-based contextual bandit baselines, with robustness validated across varying blockage rates and network configurations.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Zhou2026d,

title = {MPDS-SAC: Multi-horizon Post-Decision State Soft Actor-Critic for Air-Ground Collaborative Sensing},

author = {Quanxi Zhou and Wencan Mao and Yunfei Chen and Junru Li and Jiachen Gong and Kensuke Fukuda and Manabu Tsukada},

year  = {2026},

date = {2026-12-01},

urldate = {2026-12-01},

booktitle = {21st Asian Internet Engineering Conference (AINTEC 2026)},

address = {Taipei, Taiwan},

abstract = {Air-ground collaborative sensing, which utilizes unmanned aerial vehicles (UAVs) to collect data from ground wireless sensors, facilitates smart-city monitoring benefiting from mobility, scalability, and fast responses. However, existing model-free trajectory-planning policies learn action consequences purely from sampled experience, without exploiting environmental and scenario-specific patterns, leading to slow convergence and short-sighted decision-making. This paper proposes a Multi-horizon Post-Decision State Soft Actor-Critic (MPDS-SAC) algorithm for jointly optimizing multi-UAV trajectories in air-ground collaborative sensing. Within MPDS-SAC, SAC provides the deep reinforcement learning (DRL) backbone for trajectory optimization, while MPDS exploits deterministic transition rules and available observations to analytically predict the states resulting from two successive decisions without additional environment interactions. These additional post-decision states provide the critic with richer training signals, improving sample efficiency and value-estimation accuracy. Experiments on a real-world urban topology show gains over state-of-the-art DRL baselines in reward, convergence speed, and stability, with consistent improvements in information freshness and uplink quality as well as comparable sensor coverage.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Li2026e,

title = {GeoFlex: Geographic-Prior Flexible Adaptation of Per-Domain Diffusion Trajectory Planners to Held-Out Deployment Regions},

author = {Yun Li and Ehsan Javanmardi and Simon Thompson and Jiachen Gong and Qunli Zhang and Zifan Zeng and Shiming Liu and Peng Wang and Yidu Zhang and Manabu Tsukada},

year  = {2026},

date = {2026-11-09},

urldate = {2026-11-09},

booktitle = {Conference on Robot Learning (CoRL 2026)},

address = {Austin, Texas, USA},

abstract = {Cross-region deployment of a structured multi-domain diffusion planner reduces to a small-data fine-tuning problem rather than a per-scene routing problem. We test eight zero- or low-training routing schemes on three real-world held-out maps under a disjoint-lanelet protocol; none beats a trivial static rule that picks the geographically nearest training domain. A Fano bound and a split-conformal exchangeability rejection (p=0.0035) explain why no scene-feature router can break this static-prior ceiling, and a PAC-Bayes-certified von Mises-Fisher mixture, the strongest principled routing-class baseline, confirms the limit is architectural: any method confined to the linear span of the four per-expert weights is bounded by the static-prior fleet mean (15.85 vs. 13.74 m). Stepping outside this subspace breaks the ceiling. Joint replay, mixing target-region scenes with training scenes at 10:1 during fine-tuning, reaches 4.26 m fleet-mean on disjoint lanelets (-76% vs. frozen substrate) at +0.74 m in-region regression; the recipe replicates across all four nuPlan leave-one-city-out configurations including a -35% Singapore reduction under left-hand-traffic rule shift. As a methodological contribution we introduce variance-collapse forensics: a two-line diagnostic (within-scene K-sample variance plus cross-scene pairwise L2) that unmasks hypernetwork-conditioned LayerNorm and faithful TENT (test-time entropy minimisation) as near-stationary degenerate trajectories whose seemingly-strong fleet means would otherwise overstate the gain by an order of magnitude. The same backbone has passed real-vehicle closed-loop deployment under a production DPO LoRA (joint-replay weights are not yet on-vehicle), where adversarial-aware training lifts peak deceleration 3.4x without losing natural-data comfort.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Adhipradhana2026,

title = {DispatchRAG: Grounding Emergency Dispatch Decisions in Real-World Protocols from Traffic Accident Video},

author = {Muhammad Sulthan Adhipradhana and Ehsan Javanmardi and Naren Bao and Manabu Tsukada},

url = {https://arxiv.org/abs/2607.23132},

year  = {2026},

date = {2026-11-09},

urldate = {2026-11-09},

booktitle = {2026 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC 2026)},

address = {Bangkok, Thailand},

abstract = {Assessing the severity of a traffic accident scenario is important to decide which emergency service to dispatch. Missing an ambulance dispatch on a pedestrian accident is a fatal issue that can lead to death. Recently, Vision-Language Models (VLMs) have been a promising tool for accident reasoning, yet many VLMs are not grounded in real-life accident response protocols, making them not usable in accident severity assessment off-the-shelf. We introduced DispatchRAG, an accident assessor and dispatcher framework grounded in real-life Japanese traffic-accident response protocols, designed to enhance VLMs to generate an appropriate emergency response during an emergency scenario. Utilizing a RAG-based retrieval mechanism to retrieve the most relevant accident protocol and an LLM-powered reasoner to suggest the most proper response. To support evaluation, we introduce Accident Dispatch Dataset, a comprehensive dataset of accident assessment and emergency response according to Japanese accident response protocols adapted from the MM-AU dataset. We validate our framework on the Accident Dispatch Dataset, showing strong performance across various accident scenarios compared to the baseline VLM, pointing toward integration in autonomous vehicles that can automatically report both their own and nearby accidents.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Orsholits2026,

title = {GP-02: A Sub-50 g Head-Mounted RTK-GNSS/IMU/Barometer Pose Estimator with On-Board GNSS-Denied Recovery},

author = {Alex Orsholits and Manabu Tsukada},

year  = {2026},

date = {2026-11-06},

urldate = {2026-11-06},

booktitle = {The 4th Annual IEEE International Conference on Metaverse Computing, Networking, and Applications (MetaCom 2026)},

address = {Xi'an, China},

abstract = {Georeferenced head pose connects locally stable headset tracking to outdoor mixed-reality content, yet headset odometry ordinarily remains in a local frame and wearable GNSS provides position rather than head-referenced 6-DOF pose. GP-02 addresses this gap with a self-contained, 45.8 g head-worn device that fuses RTK-GNSS, a consumer IMU, and a barometer on an ESP32-S3. It emits georeferenced pose at a nominal 100 Hz without a phone, single-board computer, or offline processing. Because head orientation changes independently of walking direction, a selective-yaw estimator observes the head-to-travel offset from RTK course over ground and maintains the relationship during GNSS denial. Four continuous walks covering 7.26 km and 72.1 min across Tokyo characterise the system under progressively restricted sky visibility. Across the three runs with healthy orientation measurements, head and travel direction differed by more than 15 degrees for 65-89 percent of moving fixed-RTK observations. The estimated trajectories follow street geometry in open and moderately obstructed environments and continue through short GNSS outages. In the dense Nihonbashi canyon, ambiguity-float fixes and an IMU orientation fault degrade the horizontal trajectory, while barometric fusion suppresses raw vertical excursions of hundreds of metres. The results establish the operating envelope of a standalone absolute-pose wearable and identify the complementary role of headset visual-inertial odometry in dense urban environments.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}