RVT-2: Learning Precise Manipulation from Few Demonstrations

摘要

Pick and insert plugPick and insert 16mm peg Pick and insert 8mm peg Fig. 1: RVT-2 performing high precision tasks.Given a language instruction, a single RVT-2 model can perform multiple 3D manipulation tasks, including ones requiring millimeter-level precision like inserting peg in hole and inserting plug in socket.RVT-2 is trained with 10 demonstrations per task and uses only a single third-person RGB-D camera.Abstract-In this work, we study how to build a robotic system that can solve multiple 3D manipulation tasks given language instructions.To be useful in industrial and household domains, such a system should be capable of learning new tasks with few demonstrations and solving them precisely.Prior works, like PerAct [40] and RVT [17], have studied this problem, however, they often struggle with tasks requiring high precision.We study how to make them more effective, precise, and fast.Using a combination of architectural and system-level improvements, we propose RVT-2, a multitask 3D manipulation model that is 6X faster in training and 2X faster in inference than its predecessor RVT.RVT-2 achieves a new state-of-the-art on RLBench [24], improving the success rate from 65% to 82%.RVT-2 is also effective in the real world, where it can learn tasks requiring high precision, like picking up and inserting plugs, with just 10 demonstrations.Visual results, code, and trained model are provided at: https://robotic-view-transformer-2.github.io/.