On December 18, 2025, the School held a special lecture in Room 111, West Wing of Nanyong Building, featuring Dr. Shangguan Zeyu, who delivered an academic report titled Generalizing Robot Manipulation with Visual Intelligence. The lecture was chaired by Li Wenbin, a Tenure-Track Associate Professor at the School.
In the special lecture, Shangguan Zeyu shared three significant achievements. The first is the RoLA framework, which enables the transformation of any single image (such as a mobile phone photo) into an interactive physical environment through scene restoration and visual mixing technology. This approach rapidly generates large-scale, realistic robot training data without requiring additional hardware support. The second achievement is the ManipBench benchmark, a specialized evaluation standard designed to assess the low-level manipulation reasoning capabilities of Vision-Language Models (VLMs) in tasks such as object interaction and deformation handling, aiming to precisely evaluate model performance in practical operations. The third achievement involves the pick-and-place task in robotics, where a method for identifying failure precursors based on the analysis of relative object motion was proposed, enabling proactive prediction and prevention of operational failures. These three contributions hold significant scientific value in the areas of data acquisition, model evaluation, and system robustness.
The interactive session of the lecture was highly engaging, with faculty and students actively participating and raising questions related to the presentation as well as their own research concerns. This led to an in-depth and inspiring exchange with Shangguan Zeyu. With this, the lecture was successfully concluded.
