Air Series Articles Released

Air Series is a collection of articles mentored by Chen Wang.

A wide variety of topics in robotics are covered, including localization, detection, and lifelong learning.

All articles are first authored by Undergraduate or Master students and second authored by Chen Wang.

Air Series Articles

    First Author Information (When work was done)

    • Bowen Li
      • RISS intern at Carnegie Mellon University.
      • Junior student at Tongji University, China.
      • Now: Incoming PhD student of CMU RI.
    • Nikhil Varma Keetha
      • RISS intern at Carnegie Mellon University.
      • Junior student at Indian Institute of Technology Dhanbad.
      • Now: Incoming Master student of CMU RI.
    • Dasong Gao
      • Master student at Carnegie Mellon University.
      • Now: Incoming PhD student of MIT EECS.
    • Yuheng Qiu
      • Undergraduate of Chinese University of Hong Kong.
      • Now: PhD student of CMU ME.
    • Kuan Xu
      • Master Graduate of Harbin Institute of Technology, China.
      • Engineer at Tencent and Geekplus.
      • Now: Incoming PhD student of NTU EEE.

    Contribution

    • AirDet: Few-shot Detection without Fine-tunning

      • The first practical few-shot object detection method that requires no fine-tunning.
      • It achieves even better results than the exhaustively fine-tuned methods (up to 60% improvements).
      • Validated on real world sequences from DARPA Subterranean (SubT) challenge.
    Only three examples are given for novel object detection without fine-tunning.
    • AirObject: Temporal Object Embedding

      • The first temporal object embedding method.
      • It achieves the state-of-the-art performance for video object identification.
      • Robust to severe occlusion, perceptual aliasing, viewpoint shift, deformation, and scale transform.
      • Project Page
    Temporal object matching on videos.
    • AirDOS: Dynamic Object-aware SLAM (DOS) system

      • The first DOS system showing that camera pose estimation can be improved by incorporating dynamic articulated objects.
      • Establish 4-D dynamic object maps.
      • Project Page
    Dynamic Objects can correct the camera pose estimation.
    • AirLoop: Lifelong Learning for Robots

      • The first lifelong learning method for loop closure detection.
      • Model incremental improvement even after deployment.
      • Project Page
    The model is able to correct previously made mistakes after learning more environments.
    • AirCode: Robust Object Encoding

      • The first deep point-based object encoding for single image.
      • It achieves the state-of-the-art performance for object re-identification.
      • Robust to viewpoint shift, object deformation, and scale transform.
      • Project Page
    A human matching demo.

    Congratulations to the above young researchers!

    More information can be found at the research page.

    Latest Research

    AirIMU: Learning Uncertainty Propagation for Inertial Odometry
    AirIMU: Learning Uncertainty Propagation for Inertial Odometry

    AirIMU, a hybrid IO system designed for state estimation and uncertainty propagation.

    SubT-MRS: Pushing SLAM Towards All-weather Environments
    SubT-MRS: Pushing SLAM Towards All-weather Environments

    Simultaneous localization and mapping (SLAM) is a fundamental task for numerous applications such...

    TartanDrive 2.0: More Modalities and Better Infrastructure to Further Self-Supervised Learning Research in Off-Road Driving Tasks
    TartanDrive 2.0: More Modalities and Better Infrastructure to Further Self-Supervised Learning Research in Off-Road Driving Tasks