Max Robotics — Global Service Robot Directory + US Certification

About

Congyue Deng is a rising star in robotics and embodied AI, whose research focuses on building generalizable and data-efficient manipulation policies for complex, real-world tasks. Her core contributions lie at the intersection of geometric deep learning and robot learning, where she pioneers the use of equivariance—specifically SIM(3)-equivariance—to enable robots to transfer skills across objects of different scales, poses, and appearances without exhaustive retraining. Her landmark work, **EquivAct** (2024, 19 citations), demonstrates that by encoding equivariance into both visual representations and action policies, a robot can generalize from folding a kitchen towel to folding a large beach towel, a leap beyond traditional data augmentation. Deng further advances this paradigm with **EquiBot** (2024, 4 citations), a diffusion policy that achieves robust, data-efficient learning for diverse manipulation tasks. Beyond rigid objects, she tackles deformable and articulated objects: **Make a Donut** (2025, 4 citations) introduces hierarchical planning for zero-shot deformable manipulation with tools, while **NAP** (2023, 3 citations) is the first generative model for synthesizing 3D articulated objects. Her work on **PhysPart** (2024-2025, 2+ citations) addresses physically plausible part completion for interactable objects, bridging simulation and reality. With a growing citation impact and a clear trajectory toward foundational, generalizable robotic intelligence, Deng is shaping the future of how robots learn to interact with our unstructured world.