Research on Gan Luyang

Learning Open-World Visual–Tactile Grasp Stability Prediction with Synthetic Data

Mon, 01 Sep 2025 00:00:00 +0000

A model trained on a large real-world visual-tactile dataset predicts grasp success on unfamiliar objects at 53%, which is a coin flip. Trained on data we generated in simulation, the same model reaches 77.5%. Real data is supposed to be the gold standard, and the reason it loses here is not that there was too little of it.

This paper was never published, so this page is the full account of it.

FetchBench: A Simulation Benchmark for Robot Fetching

Tue, 01 Oct 2024 00:00:00 +0000

Fetching is the whole sequence: approach an object, grasp it, bring it out. FetchBench asks how well current methods do it when the object is not on a table top, and the answer is that the best baseline we assembled reaches about 20%. The paper’s contribution is not that number alone but the decomposition underneath it, which separates how much of the failure belongs to grasp prediction, to partial observation, and to execution.