Original title: Adaptive Vision-Language Grasping via Composable Foundation Priors and Generalizable Grasp Synthesis
Sixu Yan, Shikang Wang, Binhua Huang, Xuanlai Tang +10 · September 2026
Robots struggle to grasp objects when tasks change or when using different robotic hands. Many systems tie task understanding and hand control tightly together. To fix this, researchers built AdaRoboVLG. The system splits the job in two. A base policy handles the physical grasping. It maps hand movements and checks grip stability to propose workable grasps. Meanwhile, separate foundation models handle task understanding. These models provide plug-in clues about space, thinking, and time without retraining the base policy. Experiments in simulation and the real world showed that the base policy learns quickly and works across different hands. Adding these clues let robots complete functional grasps in cluttered and moving settings, matching top methods without sacrificing grasp quality or needing policy retraining.
Imagine packing a bag for a trip where you need different items for different activities. It's hard to know exactly what to bring without checking the forecast, and if you pack wrong, you might be unprepared. In this paper, the robot has a solid way to pack and carry items without dropping them, but it uses separate guides to decide what to pack based on the weather and event. This way, it can switch trips without relearning how to pack.
This makes robots more adaptable and easier to upgrade as AI improves. It benefits anyone building robots for picking tasks, from warehouses to homes, without needing to redesign the grasping system each time.
The abstract doesn't explicitly state limitations, but it shows results on three representative grasping challenges in simulation and real-world experiments.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Adaptive Vision-Language Grasping via Composable Foundation Priors and Generalizable Grasp Synthesis — Sixu Yan, Shikang Wang, Binhua Huang et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.