Selected work

Representative Papers

Awkernel: RTOS Bridging DAG Scheduling Theory and Component-Oriented Real-Time Systems Practice

Oct 04, 2026

This study addresses the lack of real hardware support in DAG scheduling theory and the difficulty existing platforms face in ensuring timing analyzability. To this end, we propose the first open-source RTOS that natively supports real-time DAG scheduling. By introducing a Function-as-Subtask (FasS) API to enforce DAG semantics and integrating a publish/subscribe model with a Global Earliest Deadline First (GEDF) algorithm, the system achieves OS-level native support for DAG task models alongside a modular scheduler with microsecond-level overhead and an automated testbed. Raspberry Pi-based experiments demonstrate that a GEDF scheduler implemented in merely 226 lines of code successfully ports autonomous driving components, enabling mechanical port migration without additional constraint reasoning. This work effectively bridges the gap between theoretical scheduling frameworks and practical engineering implementation.

0 citationsRead paper

Autoware in Construction: Gap Analysis and LiDAR Perception Toward Off-Road Autonomous Driving

Oct 03, 2026

This study addresses the challenge that unstructured construction site environments violate the foundational assumptions of conventional on-road autonomous driving systems, rendering existing frameworks inapplicable to engineering vehicles. Building upon the open-source Autoware platform, this work systematically analyzes adaptation gaps within the perception, localization, and planning modules for autonomous dump trucks, proposing an end-to-end autonomy roadmap tailored to dynamic construction environments. The research primarily reconstructs the LiDAR point cloud perception pipeline and conducts multi-module co-optimization alongside field testing. By delineating improvement pathways for each core module and validating the robustness of LiDAR perception under complex operating conditions, this study effectively bridges the technological divide between on-road and off-road scenarios, establishing both theoretical and engineering foundations for the automation of heavy-duty construction vehicles.

0 citationsRead paper

Domain Adaptation for Different Sensor Configurations in 3D Object Detection

Sep 04, 2025

To address the degradation of 3D detection performance across vehicle platforms with heterogeneous sensor configurations in autonomous driving, this paper proposes a multi-sensor-aware domain adaptation method. Our approach introduces, for the first time, a synergistic fine-tuning strategy combining downstream task adaptation and partial-layer fine-tuning, leveraging co-located, multi-sensor point cloud pairs for adaptation. Unlike conventional joint training, we selectively fine-tune only the detection head and critical backbone layers, thereby mitigating negative transfer induced by point cloud distribution shifts. Extensive experiments across diverse real-world sensor configurations demonstrate an average +4.2% improvement in BEV mAP. The method exhibits strong generalization and scalability, offering an efficient, lightweight domain adaptation solution for deploying 3D detectors across heterogeneous platforms.

0 citationsRead paper

Rethink 3D Object Detection from Physical World

Jun 30, 2025

Current 3D detection evaluation over-relies on mAP and latency as isolated metrics, neglecting the accuracy–latency trade-off, cross-hardware performance variability, and—critically—the real-world impact of detection errors on motion planning safety. To address this, we propose a novel evaluation framework introducing latency-aware AP (L-AP) and planning-aware AP (P-AP), the first metrics to explicitly quantify the coupling among detection models, hardware accelerators, and downstream planning risk. Leveraging the nuPlan dataset and latency-aware hyperparameter optimization (L-HPO), we conduct joint evaluation across diverse hardware platforms and point-cloud detectors. Experiments reveal that increased point-cloud density does not necessarily improve planning safety; our method achieves state-of-the-art real-time detection performance while providing quantifiable, hardware-aware optimization guidance for autonomous driving co-design of software and hardware.

0 citationsRead paper

AWML: An Open-Source ML-based Robotics Perception Framework to Deploy for ROS-based Autonomous Driving Software

May 31, 2025

To address low development efficiency, high annotation costs, and the absence of a closed-loop MLOps pipeline for machine learning–based perception models in ROS-based autonomous driving systems, this paper introduces the first open-source MLOps framework tailored for ROS 2. The framework innovatively integrates active learning with multimodal data mining to establish an end-to-end, closed-loop pipeline encompassing automated/semi-automated annotation, model training, ROS-native deployment, and online feedback. Built upon PyTorch and natively integrated with ROS 2, it enables reproducible, continuously evolving perception model iteration. Experimental evaluation demonstrates that the framework reduces manual annotation effort by 62% on average and accelerates model iteration cycles by 3.1×. Its deployability and robustness are validated across multiple real-world ROS-based autonomous driving scenarios.

0 citationsRead paper

Recent publications

Latest Papers

Awkernel: RTOS Bridging DAG Scheduling Theory and Component-Oriented Real-Time Systems Practice

Oct 04, 2026

This study addresses the lack of real hardware support in DAG scheduling theory and the difficulty existing platforms face in ensuring timing analyzability. To this end, we propose the first open-source RTOS that natively supports real-time DAG scheduling. By introducing a Function-as-Subtask (FasS) API to enforce DAG semantics and integrating a publish/subscribe model with a Global Earliest Deadline First (GEDF) algorithm, the system achieves OS-level native support for DAG task models alongside a modular scheduler with microsecond-level overhead and an automated testbed. Raspberry Pi-based experiments demonstrate that a GEDF scheduler implemented in merely 226 lines of code successfully ports autonomous driving components, enabling mechanical port migration without additional constraint reasoning. This work effectively bridges the gap between theoretical scheduling frameworks and practical engineering implementation.

0 citationsRead paper

Autoware in Construction: Gap Analysis and LiDAR Perception Toward Off-Road Autonomous Driving

Oct 03, 2026

This study addresses the challenge that unstructured construction site environments violate the foundational assumptions of conventional on-road autonomous driving systems, rendering existing frameworks inapplicable to engineering vehicles. Building upon the open-source Autoware platform, this work systematically analyzes adaptation gaps within the perception, localization, and planning modules for autonomous dump trucks, proposing an end-to-end autonomy roadmap tailored to dynamic construction environments. The research primarily reconstructs the LiDAR point cloud perception pipeline and conducts multi-module co-optimization alongside field testing. By delineating improvement pathways for each core module and validating the robustness of LiDAR perception under complex operating conditions, this study effectively bridges the technological divide between on-road and off-road scenarios, establishing both theoretical and engineering foundations for the automation of heavy-duty construction vehicles.

0 citationsRead paper

Domain Adaptation for Different Sensor Configurations in 3D Object Detection

Sep 04, 2025

To address the degradation of 3D detection performance across vehicle platforms with heterogeneous sensor configurations in autonomous driving, this paper proposes a multi-sensor-aware domain adaptation method. Our approach introduces, for the first time, a synergistic fine-tuning strategy combining downstream task adaptation and partial-layer fine-tuning, leveraging co-located, multi-sensor point cloud pairs for adaptation. Unlike conventional joint training, we selectively fine-tune only the detection head and critical backbone layers, thereby mitigating negative transfer induced by point cloud distribution shifts. Extensive experiments across diverse real-world sensor configurations demonstrate an average +4.2% improvement in BEV mAP. The method exhibits strong generalization and scalability, offering an efficient, lightweight domain adaptation solution for deploying 3D detectors across heterogeneous platforms.

0 citationsRead paper

Rethink 3D Object Detection from Physical World

Jun 30, 2025

Current 3D detection evaluation over-relies on mAP and latency as isolated metrics, neglecting the accuracy–latency trade-off, cross-hardware performance variability, and—critically—the real-world impact of detection errors on motion planning safety. To address this, we propose a novel evaluation framework introducing latency-aware AP (L-AP) and planning-aware AP (P-AP), the first metrics to explicitly quantify the coupling among detection models, hardware accelerators, and downstream planning risk. Leveraging the nuPlan dataset and latency-aware hyperparameter optimization (L-HPO), we conduct joint evaluation across diverse hardware platforms and point-cloud detectors. Experiments reveal that increased point-cloud density does not necessarily improve planning safety; our method achieves state-of-the-art real-time detection performance while providing quantifiable, hardware-aware optimization guidance for autonomous driving co-design of software and hardware.

0 citationsRead paper

AWML: An Open-Source ML-based Robotics Perception Framework to Deploy for ROS-based Autonomous Driving Software

May 31, 2025

To address low development efficiency, high annotation costs, and the absence of a closed-loop MLOps pipeline for machine learning–based perception models in ROS-based autonomous driving systems, this paper introduces the first open-source MLOps framework tailored for ROS 2. The framework innovatively integrates active learning with multimodal data mining to establish an end-to-end, closed-loop pipeline encompassing automated/semi-automated annotation, model training, ROS-native deployment, and online feedback. Built upon PyTorch and natively integrated with ROS 2, it enables reproducible, continuously evolving perception model iteration. Experimental evaluation demonstrates that the framework reduces manual annotation effort by 62% on average and accelerates model iteration cycles by 3.1×. Its deployability and robustness are validated across multiple real-world ROS-based autonomous driving scenarios.

0 citationsRead paper