Jian-xiang Feng / 冯健祥

Hi there! I received my PhD at Technical University of Munich (TUM) and the Institute of Robotic and Mechtronic (RM), German Aerospace Center (DLR). I was advised by Prof. Rudolph Triebel (DLR&KIT) and affiliated with Munich School of Data Science, where I worked with Prof. Stephan Günnemann. Recently, I am a senior research scientist at Agile Robots SE, investigating learning-based manipulation, e.g., vision-language-action models (VLAs), dexterous grapsing, etc..

In general, my research interests reside in the intersection of robotics and machine learning. A primary focus is on the trustworthy and adaptable learning ability of a robot in an open-world environment. Aiming to equip an autonomous agent with introspective capabilities i.e., uncertainty quantification/awareness of system internal states, I am passionate about investigating probabilistic machine learning on how to properly model/quantify it and leverage such information for learning-enabled robotics.

P.S. The pronunciation of my first name "Jianxiang" is quite close to "Jensen" (yep, the well-known "Jensen inequality" often used in the Evidence Lower Bound (ELBO) derivation) but with a different ending :P.

Short Bio     CV      Github     G. Scholar      LinkedIn    

jianxiang.feng at tum dot de

An efficient way to distill view-consistent 2D features onto 3D points based on the proposed language-enhanced feature fusion strategy, thereby enabling single-view few-shot generalization.

A novel normalizing flows-based variational approach to tackle the latent feature collapse in cVAE for diverse and efficient Grasp Synthesis. The exact flow likelihood computation enables our model to quantify the shape uncertainty of partial point clouds and detect novel geometry.

A novel way to equip NFs with efficient but flexible base distributions to overcome the topological constraint for OOD detection in robot learning.

A density-based method with normalizing flows for feasibility learning in Robotic Assembly based on only feasible examples.

A holistic graphical approach including a graph representation for product assemblies and a policy architecture, Graph Assembly Processing Network, dubbed GRACE to predict assembly sequences in a step-by-step manner.

An active learning pipeline to reduce annotation efforts of real data within a Sim-to-Real scenario based on deep Bayesian Neural Networks.

A probabilistic framework to obtain both reliable and fast uncertainty estimates for predictions with Deep Neural Networks (DNNs) based on Sparse Gaussian Processes.

A method for adaptive image classification based on fusing uncertainty estimates from Bayesian Neural Networks as unary potentials within a Conditional Random Field (CRF).

A highly integrated assistive robot EDAN, operated by an interface based on bioelectrical signals, combined with shared control and a whole-body coordination of the entire system, through a case study involving people with motor impairments to accomplish real-world activities..

A language-guided object-centric diffusion policy that takes a 3d representation of task-relevant objects as conditional input and can be guided by cost function for collision avoidance at inference time.

A Conditional Generative Adversarial Networks (cGANs)-based DexGenerator to generate dexterous grasps and a discriminator-like DexEvalautor to assess the stability of these grasps.

A dynamic grasping framework for unknown objects in this work, which uses a five-fingered hand with visual servo control and can compensate for external disturbances.

An investigation of three different ways for uncertainty quantification of a VLM failure detector for closed-loop LLM planners.

A comprehensive overview of uncertainty estimation in neural networks, reviewing recent advances in the field, highlighting current challenges, and identifying potential research opportunities.

A novel teleoperation system for advancing aerial manipulation in dynamic and unstructured environments based on pose estimation pipelines for the industrial objects of both known and unknown geometries and an active learning pipeline.

A sparse representation of model uncertainty for Deep Neural Networks (DNNs) where the parameter posterior is approximated with an inverse formulation of the Multivariate Normal Distribution (MND), also known as the information form.

Academic Services

• International Conference on Learning Representations (ICLR) 2025
• Conference on Robot Learning (CoRL) 2022-2024
• IEEE International Conference on Robotics and Automation (ICRA) 2020&2022&2025
• IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2020&2022&2025
• European Conference on Artificial Intelligence (ECAI) 2020
• IEEE-RAS International Conference on Humanoid Robots 2024
• IEEE Transactions on Circuits and Systems for Video Technology (TCSVT)
• IEEE Robotics and Automation Letters (RAL)

• ICRA2024 Workshop on Back to the Future: Robot Learning Going Probabilistic
• IROS2022 Workshop on Probabilistic Robotics in the age of Deep Learning

Events

Mentorship

Fun Moments and Hobbies

Link to the template. Do not scrape the HTML from this page itself, as it includes analytics tags that you do not want on your own website — use the github code instead. Also, consider using Leonid Keselman's Jekyll fork of this page.