13 June 2024 in IEEE Robotics and Automation Letters
In the context of autonomous driving, it is both critical and challenging to locate 3D objects by using a calibrated RGB image. Current methods typically utilize heteroscedastic aleatoric uncertainty loss to regress the depth of objects, thereby reducing the impact of noisy input while also ensuring the reliability of depth predictions. However, experimentation reveals that uncertainty loss can also lead to serious overfitting issue and performance degradation. To address this issue, we propose MonoSample, an augmentation method that collects samples from the dataset and places them randomly during training. MonoSample takes into account the occlusion relationships and applies strict restrictions to ensure the verisimilitude of the enhanced scenes. Furthermore, MonoSample avoids the complex conversion process between 2D and 3D, thereby enabling the extraction of a large number of samples and efficient operation. Experiments on different models have verified its effectiveness. Leveraging MonoSample in DID-M3D, our model achieves state-of-the-art (SOTA) performance on the KITTI 3D object detection benchmark. The code will be published in https://github.com/EE615/MonoSample .
17 October 2025 in IEEE Robotics and Automation Letters
Autonomous driving algorithms rely on high-fidelity closed-loop simulators for safety validation. However, existing scene reconstruction techniques exhibit ghosting artifacts and unseen-view distortion at off-trajectory viewpoints, constraining navigable areas in simulations. To address this limitation, we propose ConsistentGS—a novel driving scene reconstruction framework aimed at expanding drivable regions. Our approach progressively introduces pseudo-views via raycast-based sampling and refines degraded renderings using a diffusion-based restoration network. Directly integrating pseudo-views causes conflicts and haze-like artifacts. We resolve this by introducing a gradient gating module conditioned on multi-view consistency. This mechanism selectively suppresses gradients in certain regions while preserving learning in problematic areas. Building upon OmniRe, ConsistentGS significantly enhances off-trajectory fidelity in driving scenes, enabling closed-loop simulation across expanded navigable regions. Extensive evaluations on the Waymo dataset show that our method outperforms state-of-the-art approaches. Notably, ConsistentGS achieves a 50% reduction in average FID and a 1.2 dB improvement in PSNR over baselines. The code will be open-sourced upon publication.
16 October 2026 in Measurement Science and Technology
Connected autonomous vehicles (CAVs) can improve target tracking by sharing distributed measurements of surrounding traffic participants. However, cooperative measurement fusion is reliable only when both measurement uncertainty and inter-vehicle dependence are properly modelled. Sequential Kalman filtering provides a communication-efficient late-fusion solution for cooperative target tracking, but its standard formulation relies on statistically independent and temporally aligned measurements. In practical connected-vehicle systems, this assumption can be violated by shared detection algorithms, common environmental conditions, and communication delay. As a result, correlated or temporally misaligned measurements may be repeatedly counted, leading to underestimated covariance and overconfident target-state estimates. This paper proposes ReCoTrack, a correlation-aware measurement fusion framework for reliable cooperative 3D target tracking. ReCoTrack preserves the association advantages of sequential Kalman filtering while selectively applying covariance intersection (CI) to targets jointly observed by multiple vehicles. This design enables conservative state fusion under unknown inter-vehicle measurement correlations. To improve uncertainty propagation, a trajectory-based process-noise estimator is further introduced to learn motion-dependent covariance from historical target states. An innovation-consistency objective is also designed to jointly calibrate observation and process noise covariances. Experiments on the V2V4Real dataset show that the proposed method improves Average Multi-Object Tracking Accuracy (AMOTA) by 2.8%, increases mostly tracked trajectories by 4.5%, and reduces mostly lost trajectories by 10.9% under the same late-fusion communication setting, while evaluation on OPV2V under a unified tracking protocol shows that ReCoTrack achieves the highest AMOTA among the compared methods and outperforms DMSTrack by 1.05 percentage points. The framework requires only 0.0073 MB per CAV per frame. Additional tests under controlled cross-CAV dependence and stochastic communication degradation demonstrate that the proposed method improves both tracking accuracy and covariance reliability.
22 May 2026 in Neurocomputing