Junnan Jiang (蒋俊南)

Email: elevenjiang8 [at] gmail [dot] com

Github: https://github.com/elevenjiang1

Research Interests: Robotic Grasping, Computer Vision, Robotic Manipulation

• Cleaning robot with robot arm.
• Data generation.
• Large language model.

• Eye-on-Hand camera for internal wiring of robotic arms.
• Configuration based programming for 2D vision
• Basic visual 2D and 3D tasks, hand-eye calibration, plane alignment, spatial point selection, workpiece 6D pose estimate, etc.
• Basic QT code for delivering projects.

• Responsible for the whole development process of the chess robot. Including system calibration, chessboard recognition, robot arm control, embedding development. And also participate in exhibition plaform interaction design, front-end UI display, chess algorithm deployment.
• Use EMG signal to control dexterous hand (ShadowHand, AllegroHand) to play the piano.

• A real-world dataset is collected based on teleoperation, using Shadow hand equipped with tactile sensor of BioTac. And a novel PoseFusion network is proposed to select the three estimated pose candidates and thus can well address the problem of corrupted data.
• Paper "PoseFusion: Robust Object-in-Hand Pose Estimation with SelectRNN" Yuyang Tu, Junnan Jiang, Shuang Li, Norman Hendrich, Miao Li and Jianwei Zhang, (Accepted by IROS, co-first author).

• Generate new data that leverage the feature of original high rarity and graspness shapes, which are both rare in a grapsing dataset, to improve the quality of the grasping dataset.
• Paper "Improving Robotic Grasping Ability Through Deep Shape Generation" Junnan Jiang, Yuyang Tu, Xiaohui Xiao, Zhongtao Fu, Jianwei Zhang, Fei Chen, Miao Li, (Accepted by ICIRA).

• The approach leverages teleoperation for data collection, adopts a sampling-based method for imitation learning, and utilizes a Masked AutoEncoder for multimodal fusion encoding.
• Paper "Learning Autonomous Ultrasound via Latent Task Representation and Robotic Skills Adaptation" Xutian Deng, Junnan Jiang, Wen Cheng, Miao Li (Accepted by RSS workshop).

• Using contrastive learning techniques for data transfer at the grasping feature level, enabling grasping ability transfer across different scenes. The main benefit of this method is that it is no need to re-label grasping data in the new scenes.
• Paper "GraspAda: Deep Grasp Adaptation through Domain Transfer" Yiting Chen, Junnan Jiang, Ruiqi Lei, Yasemin Bekiroglu, Fei Chen, Miao Li, (Accepted by ICRA).

• Responsible for the whole field recognition of the radar station, uses binocular cameras to detect and locate robots in the field, and develops semi-automatic labeling software, using tracking, background subtraction and other methods for data labeling.
• Responsible for the autonomous moveing of engineering vehicles, use a depth camera to detect resource stations in the field, and control the vehicles to move autonomously to the resource stations.
• Coordinate the barrel improvement task, upgrade the robot mechanism with mechanical and embedded students, and capture the projectile launch distribution with a high frame rate camera, and gain experience in machining methods and project schedule management.